Int. J.

Human-Computer Studies 64 (2006) 10711083

Evaluating the consistency of immediate aesthetic
perceptions of web pages
Noam Tractinsky
a,
, Avivit Cokhavi
a
, Moti Kirschenbaum
a
, Tal Shar
b
a
Department of Information Systems Engineering, Ben-Gurion University of the Negev, Beer Sheva, Israel
b
Department of Industrial Engineering and Management, Ben-Gurion University of the Negev, Beer Sheva, Israel
Received 23 August 2005; received in revised form 5 June 2006; accepted 13 June 2006
Communicated by P. Zhang
Available online 23 August 2006
Abstract
Two experiments were designed to replicate and extend [Lindgaard et al.s, 2006. Attention web designers: you have 50 ms to make a
good rst impression! Behaviour and Information Technology 25(2), 115126] ndings that users can form immediate aesthetic
impression of web pages, and that these impressions are highly stable. Using explicit (subjective evaluations) and implicit (response
latency) measures, the experiments demonstrated that, averaged over users, immediate aesthetic impressions of web pages are remarkably
consistent. In Experiment 1, 40 participants evaluated 50 web pages in two phases. The average attractiveness ratings of web pages after a
very short exposure of 500 ms were highly correlated with average attractiveness ratings after an exposure of 10 s. Extreme attractiveness
evaluations (both positive and negative) were faster than moderate evaluations, landing convergent evidence to the hypothesis of
immediate impression. The ndings also suggest considerable individual differences in evaluations and in the consistency of those
evaluations. In Experiment 2, 24 of the 50 web pages from Experiment 1 were evaluated again for their attractiveness after 500 ms
exposure. Subsequently, users evaluated the design of the web pages on the dimensions of classical and expressive aesthetics. The results
showed high correlation between attractiveness ratings from Experiments 1 and 2. In addition, it appears that low attractiveness is
associated mainly with very low ratings of expressive aesthetics. Overall, the results provide direct evidence in support of the premise that
aesthetic impressions of web pages are formed quickly. Indirectly, these results also suggest that visual aesthetics plays an important role
in users evaluations of the IT artifact and in their attitudes toward interactive systems.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Web-page design; Aesthetic perceptions; Attractiveness; Classical aesthetics; Expressive aesthetics; Response latency; Response time; First
impression
1. Introduction
First impressions colour subsequent search for informa-
tion and sway judgement and choice processes. One of the
most notable sources of rst impressions is visual
appearance. In a seminal paper, Dion et al. (1972),
demonstrated that a persons physical appearance inu-
ences other aspects of the social interaction. Subsequent
studies showed how prevalent and powerful this phenom-
enon is. For example, beautiful people earn more on the
marketplace (Hamermesh and Biddle, 1994), and better
looking university instructors receive higher teaching
evaluations (Hamermesh and Parker, 2005). Clearly, effects
of visual appearance are not limited to perceptions of
humans. We are affected by the aesthetics of nature and of
architecture (e.g. Nasar, 1988a; Porteous, 1996) as well as
the beauty of everyday objects and artifacts (Postrel, 2002;
Coates, 2003; Norman, 2004a). There is also growing
evidence suggesting that evaluations of interactive systems
are inuenced in general by the systems visual appearance
(Tractinsky et al., 2000) and by the appearance of web
pages in particular (Karvonen, 2000; Schenkman and
Jonsson, 2000; Zhang and von Dran, 2000; Zhang et al.,
2001; van der Heijden, 2003; Kim et al., 2003). However, it
is not clear whether this inuence stems from immediate,
ARTICLE IN PRESS
www.elsevier.com/locate/ijhcs
1071-5819/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhcs.2006.06.009

Corresponding author. Tel.: +972 7 6472226; fax: +972 7 6472958.

E-mail address: noamt@bgu.ac.il (N. Tractinsky).
rst impression or from a more elaborated examination of
the web site. In fact some even doubt that such inuence
even exists (Hassenzahl, 2004). Thus, the causal relation-
ship between aesthetic perceptions and other perceived
attributes of the IT artifact cannot be taken for granted.
One of the reasons for the major role of aesthetics in
everyday life may stem from its immediate effect on our
senses and, consequently on our judgement. However, there
are two prerequisites to this possible explanation. Firstly, it
should be demonstrated that aesthetic impressions could be
made quickly. Secondly, the validity of this argument may
diminish if those impressions were short-lived and transient.
In this case, initial aesthetic impressions will be swept aside
by new ones, and more elaborated aesthetic evaluations
may change the overall evaluations. Thus, it is important to
demonstrate empirically that rst aesthetic impressions are
not only quick, but that they are also lasting and
consistent.
1
Such demonstration lies at the heart of this
study. Our major objective is to demonstrate the immediacy
and the consistency of aesthetic impressions.
In a series of experiments, Lindgaard et al. (2006)
provided evidence in support of both prerequisites in the
context of users evaluations of web pages. In the central
experiment conducted by Lindgaard et al. (2006, Study 2),
participants watched images of 50 web pages, each for a
brief exposure of 500 ms. This exposure time was intended
to be long enough to form a rst impression, yet not
sufciently long to evaluate other features of the web site,
such as its semantic content. After each page was shown
the participants rated its visual attractiveness
2
by using a
continuous rating scale, ranging from 0 (for very unat-
tractive web pages) to 100 (for very attractive web pages).
Then, each participant viewed the 50 pages for a second
time in a newly randomized order. The correlation between
the mean evaluation of the visual attractiveness of web
pages in the rst phase and the mean evaluations in the
second phase was 0.97, indicating that, when aggregating
individual evaluations, even very short exposure resulted in
remarkably consistent aesthetic evaluation. In a subsequent
experiment (Study 3 in Lindgaard et al., 2006), exposure
time was limited to 50 ms in an attempt to demonstrate that
evaluating web pages may follow the pattern of mere
exposure effect (Zajonc, 1980; Bornstein, 1992). The
aggregated individual evaluations under the 50 ms condi-
tion were still very consistent, but intra-user consistency
was considerably lower than in the 0.5 s condition.
Lindgaard et al. speculate that this difference may have
stemmed from users being able to take in more page
content during the 500 ms condition than in the 50 ms
condition.
This study was designed to replicate, extend, and
augment the novel and important ndings of Lindgaard
et al.s study. In the experiments reported herein we have
concentrated on the results of the 0.5 s condition and not
on the mere exposure effect experiment, although our
ndings may shed some light on this issue as well.
Replication and extension research, especially by indepen-
dent researchers (i.e. not the original researchers reporting
new ndings) serves an important function in the
advancement of science (Ehrenberg, 1990; Hubbard and
Armstrong, 1994) by helping to evaluate the validity and
the generalizability of previous studies. We elaborate on
the elements of replication and extension in Section 2.
Sections 3 and 4 present two experiments that were
conducted to accomplish the studys objectives.
2. Study objectives
As mentioned above, the major objective of this study is
to demonstrate the immediacy and the consistency of
aesthetic impressions. This will be done in large part by
replicating and extending the ndings of Lindgaard et al.
(2006). This process includes four sub-goals:
1. Generalization: Ehrenberg (1990) argues that in order to
face the realities of real-world data we must always be
looking to see whether there is a generalizable result that
holds across many different sets of data (p. 196). Thus,
the studys most fundamental objective is to generalize
Lindgaard et al.s (2006) ndings by using a new set of
data that is based on different web pages, a different
rating scale and with participants from a different
culture. The specic details of the characteristics of our
data sets are presented in the sections that describe the
two experiments. Beyond generalization, the study has
three additional goals:
2. Convergence: Here, we attempt to validate the results
using converging evidence from an implicit measure
(response latency) in addition to the use of an explicit
measure (subjective rating).
3. Comparison of two types of evaluation consistency: The
rst type refers to the consistency by which evaluations
of web pages are averaged over users in the two rounds
of evaluations. However, it is also of interest to study a
second type of consistency, one that gauges the degree to
which individuals are internally consistent in evaluating
aesthetic stimuli (e.g. Hassenzahl, 2004) and in parti-
cular stimuli that were presented for a very short
duration.
4. Association of design dimensions and attractiveness of
web pages: Here, we examine the relations between the
attractiveness of web pages and two perceived dimen-
sions of web-page designclassical aesthetics and
expressive aesthetics (Lavie and Tractinsky, 2004).
We elaborate on the last three goals below.
ARTICLE IN PRESS
1
We do not suggest that initial aesthetic impressions never change upon
further reections or experience. Our claim is that, in general, those
impressions last considerable time.
2
In this work, we use the term attractiveness following its use by
Lindgaard et al. (2006). In essence, we use this term interchangeably with
visually pleasing or (visual) aesthetics perceptions or e.g. Lavie and
Tractinsky (2004).
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1072
2.1. Response latency
Lindgaard et al. (2006, Study 2) used a single-item
subjective measure to study whether very short exposure
can elicit aesthetic response. Naturally, the results obtained
by such measures can be susceptible to measurement bias
and error. In this study we use, in addition to Lindgaard
et al.s measure, an objective measure, response latency.
Convergence of the results obtained from these two
different measures will substantially increase the ndings
validity. Response latency is the length of time taken by a
respondent to answer a question. As a measure, response
latency has several general advantages: it is unobtrusive
and is very easy to collect over computerized systems. In
addition, there are three specic reasons for our interest in
this measure.
1. Our major thesis is that certain aesthetic responses are
immediate and precede elaborated cognitive processes.
Obviously, if the process of interest involves chronometric
considerations, the appropriate measure of choice should
be one that measures durations. Indeed, research on the
immediacy of affective responses, on the spontaneity of
judgement and on the automaticity of information proces-
sing has made considerable use of response latency data
(e.g. Herr and Page, 2004). For example, by measuring
response latencies, Duckworth et al. (2002) have shown
that evaluative response to both novel and familiar stimuli
can be immediate, unintentional, and appropriate (in the
sense that the response matches the stimulus message).
2. Research suggests that response latency can be used as
a measure of strength of preferences between alternatives
(e.g. MacLachlan et al., 1979; Tyebjee, 1979; Aaker et al.,
1980). Response latency is also indicative of implicit
attitudes towards objects (e.g. Greenwald et al., 1998)
and of the strength of attitudes (e.g. Bassili, 1996).
Research on consumer behaviour suggests that the time
needed to choose between two brands is inversely propor-
tional to the psychological distance, which separates the
two brands. This is because if the brands appear to be more
similar, the choice will be more difcult, and hence take
longer, than if one brand clearly dominates the others
(Tyebjee, 1979, p. 96). MacLachlan et al. (1979) suggest
that the faster an answer is given, the stronger the
respondents conviction (p. 573). This effect has been
demonstrated regardless of whether the questions were
phrased in positive or negative terms (Shipley et al., 1946).
The effect was also demonstrated regardless of whether
people replied to factual items or expressed subjective
preferences. Thus, in MacLachlan et al. (1979), partici-
pants were asked about the makers of certain cereal
brands. Average response time for correct answers was
1.5 s faster than for wrong answers. Dashiell (1937) showed
subjects pairs of colours and asked what colour was
preferred. The stronger a colour was preferred, the faster it
was chosen. In addition, it was found that attitudes
expressed in shorter response latencies predict greater
resistance to persuasion (Stocke , 2003) and that behaviour-
al intentions (e.g. voting) expressed in shorter latencies
were better predictors of actual behaviour (Bassili, 1996).
Finally, conicting thoughts or feelings take longer to
express than coherent ones (Bassili, 1996).
Most studies employing response latency measures to
infer preferences have used binary choice tasks in which the
psychological distance between the two stimuli was
manipulated and its effects on response latencies observed.
Because of the large space required to present web pages
for users evaluations, our study is limited to registering
latencies from judgements of stimuli (i.e. a web page) that
are presented one-by-one. Relatively little research has
been done using response latencies in this mode. However,
it is reasonable to propose that the more extreme
evaluations of web-page attractiveness will be associated
with shorter response latencies. This proposition is based
on two lines of reasoning. Firstly, Ostrom and Gannon
(1996) found that extreme ratings on an evaluation scale
are easier to generate than are ratings at the middle of the
scale. Hence, we believe that generating extreme ratings
should be shorter than nonextreme ratings. Secondly,
based on the studies surveyed in this sub-section, users
conviction will be stronger when they judge pages that are
considerably more or considerably less attractive than
average pages. Presumably, very attractive or very un-
attractive web pages represent a coherent object towards
which feelings are manifested quickly. Web pages that are
not decidedly attractive or unattractive are probably not
perceived coherently from an aesthetic viewpoint and
therefore their evaluations will be slower due to the
inconsistent feelings towards them.
Indeed, Pham et al. (2001) found that extreme ratings of
pictureswhether positive or negativewere associated
with lower latencies than were more moderate ratings.
Bassili (1996) reports similar pattern of the relations
between response latency and extremity of opinion.
In the context of aesthetic evaluations of furniture,
Ritterfeld (2002) found that response latency was shorter
for prototypical stimuli (i.e. stimuli that already conform
to a preference schema) than for polyvalent stimuli
(i.e. stimuli that involve greater level of preference
uncertainty). In addition, Ritterfeld (2002) predicts that
latencies of very negative evaluations will be shorter than
latencies of very positive evaluations, because negative
evidence looms larger during the decision-making process.
For example, in the context of HCI, users may look for
additional evidence before deciding that a web page is
perfect, whereas a few salient negative features may sufce
to conclude that a web page is totally unattractive.
3. Finally, Aaker et al. (1980) suggest that measuring
response latency can be helpful in assessing the construct
validity of preference measurements. The idea is that a
true test of construct validity requires maximally
different methods to determine convergent and discrimi-
nant validity (p. 237). As a very different measurement
method relative to explicit ratings of stimuli, response
latency can serve as such a method. Thus, the use of the
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1073
two methods (rating and response latency) can serve as a
test for convergence across different measures of the same
thing (Cook and Campbell, 1979), given that, as
mentioned above, we expect response latency to covary
with rating extremity.
2.2. Evaluation consistency
As mentioned above, in order to support the premise
that aesthetic perceptions of the IT artifact (in our case,
web pages) inuence subsequent judgements of the artifact,
it is necessary to demonstrate that these perceptions remain
stable (i.e. consistent) over time. The consistency of web-
page aesthetics can be assessed in two different ways: (1)
Users evaluations of web pages can be averaged for each
page. The association between averages of repeated
evaluations can then be indicative of the degree to which
the studys sample (and by generalization the target
population) maintains a consistent rating of each page
relative to other pages. (2) The evaluations of the web
pages are compared to repeated evaluations of the same
pages within individual users. This method gauges the
degree to which individuals are internally consistent in
evaluating aesthetic stimuli (e.g. Hassenzahl, 2004). Ac-
cording to Monk (2004), for HCI designerswho design
for populations rather than for individual usersthe
interesting question is whether, on average, products are
rated consistently higher or lower relative to other
products. However, it is also of interest to study the
second type of consistency, that is, to assess the stability of
aesthetic perceptions within individuals. Lindgaard et al.
(2006) found that for very short stimuli exposure (0.5 s)
sample-averaged attractiveness estimations were consider-
ably more consistent than estimations of individual users.
This difference may be due to the fact that the error
associated with single estimations is reduced when inde-
pendent estimations (i.e. from different, independent
raters) are averaged. Lindgaard et al. (2006) examined
intra-individual consistency based on correlations of each
individuals ratings with the average rating of the entire
sample. We assess consistency in this study by correlating
participants own ratings of the web pages on two
occasionsthe rst after a 0.5 s exposure and the second
after a 10 s exposure.
2.3. Aesthetic dimensions of web-page design
We are also interested in analysing how design affects
attractiveness. In other words, we were looking for design
factors that are associated with attractive or unattractive
web pages. The design of web pages involve many
technical, low-level aspects that may also interact with
each other, yielding a virtually innite number of
components to consider when analysing the sites design.
It is therefore essential to consider higher-level, more
abstract concepts of web-site design. The HCI literature
has yet to generate considerable research of such design
concepts. First steps in this direction have been taken by
Kim and his associates (Kim et al., 2003; Park et al., 2004)
who identied design factors that relate to web-page
objects, backgrounds and the relationships between the
objects and the backgrounds. Kim et al.s (2003) work,
however, does not allow for an overall score for any of the
higher-level design factors, while Park et al.s (2004) work
identied 13 design factors, which may indicate insufcient
level of abstraction for the purpose of this study.
Lindgaard et al. (2006) have also attempted to associate
several design factors with overall assessment of web-page
attractiveness. However, it seems that their design factors
were chosen ad hoc and the results failed to distinguish
between the contributions of each factor to overall
attractiveness evaluations.
In another line of research, Lavie and Tractinsky (2004)
proposed that people identify two high-level aesthetic-
related dimensions of web pages. The dimensions
classical aesthetics and expressive aestheticsrepresent
users evaluation of the design. Classical aesthetics refers
to the orderliness and clarity of the design. Expressive
aesthetics refers to the originality, creativity and the
richness of the design. While the two dimensions are based
on users subjective evaluations of web pages, they seem to
represent general and consensusal notions of aesthetics.
For example, the two dimensions, which emerged in the
context of web pages, closely resemble the aesthetic
dimensions that emerged in works in other elds, e.g.
environmental aesthetics (Kaplan, 1988; Nasar, 1988b) and
design (Coates, 2003). The dimensions proposed by
Lavie and Tractinsky (2004) resemble aesthetic notions
such as concinnity and novelty (Coates, 2003). Classical
aesthetic represent order and familiarity, whereas expres-
sive aesthetics represent noveltytwo visual aspects of
the environment that induce pleasure. Indeed, Lavie
and Tractinsky found that both dimensions are positively
correlated with pleasing interaction, but that classical
aesthetics is also strongly correlated with perceived
usability. However, the above-mentioned correlates of
the two aesthetics dimensions represent constructs that
appear to stem from reection, whereas attractiveness
represents a more visceral response. Thus, it is difcult to
predict which of the aesthetic dimensions is more strongly
correlated with the attractiveness of web pages. Based on
the similar correlations with the pleasure construct, as
reported by Lavie and Tractinsky (2004), we expect
that both dimensions will be correlated similarly with
attractiveness.
To summarize this section, we propose to study the
immediacy and the consistency of aesthetic impressions.
The study will be based on replicating the ndings of
Lindgaard et al. (2006); on extending that research with an
implicit, independent measure; and on examining what
high-level design dimensions correlate with immediate
attractiveness evaluations. The study is comprised of two
experiments. Experiment 1 was designed to answer the rst
three goals stated at the beginning of this section.
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1074
Experiment 2 was designed to answer the fourth goal as
well as to provide additional evidence for the rst goal.
3. Experiment 1
3.1. Method
3.1.1. Sample
Forty undergraduate business students (25 female, 15
male), enrolled in decision making and organizational
behaviour classes volunteered to participate in the study
for course credit. They were 1928 years old
(average 23.7). Prior to the experiment, the students
did not participate in any class related to web-page design.
3.1.2. Stimuli
Fifty web pages (as in Lindgaard et al.s studies 2 and 3)
were evaluated in this study. The 50 web pages were
arbitrarily selected for this experiment based on two
criteria: (1) They did not belong to well-known web sites
(to reduce the possible inuence of familiarity on evalua-
tions). (2) The stimulus set had to cover a wide range of
attractiveness in order to be ecologically representative.
Hence, we selected 25 pages that we considered to be
relatively attractive and 25 pages that we considered to be
relatively unattractive.
3
The web pages came from a variety
of domains (e.g. web developers, entertainment, art and
design, retail, and personal web pages). Screen shots of the
web pages were captured at a resolution of 1024 768
pixels in 24-bit true colour, but were compressed to JPG
format with a resolution of 800 600 before being
presented in the experiment. A subset of 24 web pages
used in this experiment is presented in the Appendix.
3.1.3. Procedure
Participants were briefed about the studys general
purpose and were given written instructions regarding the
experimental task. In addition, each experimental phase
was preceded by online instructions.
The participants interacted with a computer system that
included a P4 1.7 MHz processor and a 19 in display. A C#
program in .NET environment was built to control the
procedure, to present images of web sites, to control the
display time of the images and to collect user data,
including ratings of the web-site images and response
latencies. The study consisted of two main phases. In Phase
1, each web-page image was displayed for 500 ms, after
which the rating scale was displayed on the screen and the
participants were asked to rate the attractiveness of the
page that they just saw. The rating scale was represented
by 10 radio buttons arranged in order from left to right.
A Very Unattractive and a Very Attractive verbal
anchors were placed below the 1 button and the 10
button, respectively. There were no instructions or time
limits regarding speed of rating.
4
Before the experimental
stimuli, a block of 10 trial images was administered to get
the participants acquainted with the rating method and the
short display times. Next, the 50 web-page images were
presented in a random order; participants rated each image
in turn, and pressed a Continue button when they were
ready to proceed to the next image. After that phase, an
instructions page informed the participants that they are
done with the rst phase and that the second phase is about
to begin. The stimuli and the procedure in Phase 2 were
identical to those of Phase 1, with the exception that images
were presented for 10 s rather than for 0.5 s. The order of
presentation of web-page images was again randomized for
each participant.
3.2. Results
Overall, 2000 attractiveness ratings and response laten-
cies were collected in Phase 1 (40 participants 50 web
pages). As Fig. 1 shows, the distribution of attractiveness
evaluations suggests a quasi-normal distribution where
more evaluations concentrate at the middle of the scale and
fewer pages are evaluated as extremely attractive or
unattractive. Also, extreme positive ratings are even rarer
than extreme negative ratings. While the stimuli for this
experiment were selected arbitrarily, there appears to be a
reasonable distribution of attractiveness ratings, which
limits concerns that a skewed sample of stimuli might limit
ARTICLE IN PRESS
0
50
100
150
200
250
300
350
1 3 4 5 7 8 9 10
Attractiveness Ratings
F
r
e
q
u
e
n
c
y
0
10
20
30
40
50
60
70
80
90
100
C
u
m
u
l
a
t
i
v
e

P
e
r
c
e
n
t
Frequency Cumulative Percent
6 2
Fig. 1. Frequency distribution and cumulative percent distribution of
2000 attractiveness ratings in Phase 1 (1 very unattractive, 10 very
attractive).
3
The rationale behind the selection was not to select extremely attractive
and extremely unattractive pages. Rather, we wanted to cover the range of
attractiveness that one would encounter on the Web, including
intermediate levels of attractiveness. As can be seen in Section 3.2, this
objective was met.
4
Practice differs regarding the use of speed instructions between the
preference measurement paradigm (no such instructions) and the implicit
attitude measurement paradigm. In both cases, though, the results are
quite consistent regarding the relation between response latency and
attitude or preference strength.
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1075
the generalizability of the studys results (e.g. Norman,
2004b).
3.2.1. Attractiveness evaluations of web pages
Evaluations were somewhat more favourable in Phase 2
(mean 5.29, SD 1.31) compared to Phase 1
(mean 5.06, SD 1.25). This difference is signicant
according to a paired-sample t-test (t(49) 3.23,
p 0.002). In both cases, the average ratings were just
slightly below the middle of the rating scale, indicating
that the set of web pages chosen for this study was
quite balanced in terms of the pages attractiveness. I
n Phase 1, the average attractiveness of the web pages
ranged from 2.5 for the least attractive page to 7.98 for
the most attractive page. In Phase 2 the attractiveness
evaluations ranged from 3.3 to 8.3. For each participant,
the ratings of the web pages in each phase were
transformed into z-scores to control for individual rating
tendencies. The means of the raw scores and of the z-scores
for each web page were calculated separately for each of
the two experimental phases. The correlation of the
mean z-scores of the 50 pages between the two phases
was 0.92, which was practically identical to the correlation
between the mean raw ratings of visual appeal for each
web page on both phases (r 0.92). The relation between
the raw ratings in both phases is depicted in Fig. 2.
The high-explained variance (r
2
0.85) indicates that
even with minimal exposure (i.e. only 0.5 s), the evaluation
of web-page attractiveness (averaged over participants)
was very consistent. These ndings replicate the results of
Study 2 of Lindgaard et al. (2006) despite the use
of different web pages, a different sample from a different
culture, an extended exposure in the second phase of
the experiment and a slightly modied methodology
(e.g. recording participants ratings using 10 radio
buttons rather than a slider indicating a range of 0100).
3.2.2. Within-participants analyses
The mean evaluations within participants (over 50 web
pages) ranged from 3.20 to 6.66 in Phase 1 and from 2.78 to
6.94 in Phase 2. The results are depicted in Fig. 3, which
shows a noticeable variance in participants average
ratings, together with a slight tendency to rate the pages
as more attractive in the second phase relative to the rst
phase.
The correlations of web-page evaluations within parti-
cipants reect the level of each participants consistency in
rating the attractiveness of the 50 web pages in both phases
of the study. The within-subject correlations ranged from
0.09 to 0.90, with an average correlation of 0.55 and a
median of 0.60. These results indicate large variation in
individual consistency.
3.2.3. Relation between rating extremity and response
latency
The columns in Fig. 4 depict the mean and the median of
the 2000 response latencies as a function of each attrac-
tiveness rating obtained in Phase 1. It can be seen that both
the mean and the median latencies of very attractive or
very unattractive web pages are shorter than latencies of
ratings that were placed at the middle of the scale.
To test the relation between extremity of attractiveness
rating and reaction times, an ANOVA was performed with
ratings as random factors and the transformed latencies as
a dependent variable.
5
Ratings were treated as random
factors with 5 levels based on their distance from the scales
mid-points. For example, ratings of 5 and 6which are the
scales mid-pointbelong to Category 0, whereas ratings
of 1 and 10which are the most extreme ratingsbelong
to category 4. There was a signicant effect of extremity of
ARTICLE IN PRESS
2
3
4
5
6
7
8
9
2 4 6 8 9
Phase 1
P
h
a
s
e

2
7 5 3
Fig. 2. Average raw rating for each web page in Phases 1 and 2 (each dot
reects the mean rating over 40 participants of each of 50 web pages).
2
3
4
5
6
7
8
9
2 4 5 8
Average Ratings in Phase 1
A
v
e
r
a
g
e

R
a
t
i
n
g
s

i
n

P
h
a
s
e

2
9 7 6 3
Fig. 3. Average ratings by participants in Phases 1 and 2 (each dot reects
the mean rating over 50 web pages by one of the studys 40 participants).
5
A logarithmic transformation is often used to reduce the skewness of
response latency data.
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1076
the rating on response latency (F(4, 1995) 10.815,
po0.001). The weighted linear trend was signicant
(F 38.502, po0.001), while the term of deviation from
linearity was insignicant (F 1.586, p 0.191). Table 1
displays the post-hoc comparisons between response
latencies of the ve pooled rating categories. The compar-
isons support the premise that the most extreme
evaluations (Category 4) were signicantly faster than
other evaluations and that the least extreme evaluations
(Category 0) were the slowest.
Finally, we tested the difference in latencies between
extremely positive ratings (ratings of 9 and 10) and
extremely negative ratings (1 and 2). We collapsed the
ratings of 9 and 10 into one category of extreme positive
evaluations (for a total of 148 evaluations) and ratings of 1
and 2 into a category of extreme negative evaluations (317
evaluations). As expected, latencies of positive evaluations
(mean 0.66, SD 0.37) were signicantly longer than
latencies of negative evaluations (mean 0.58, SD 0.40)
[F(1, 463) 7.786, p 0.029].
4. Experiment 2
Experiment 1 established that, on average, web pages
that are perceived as attractive after a very short exposure
are also perceived as attractive after longer exposure. An
obvious question, thus, is what design characteristics
distinguish between the more and the less attractive web
pages. One may look for low level, very specic design
factors (e.g. the use of a certain colour combination or the
placement of a certain object in a certain place). Such
research is still in its infancy (e.g. Park et al., 2005) and
solid ndings were not available when this study was
conducted. Alternatively, one can look for higher level,
more general characteristics. In this study, we used two
dimensions that represent the aesthetics of web-page design
(Lavie and Tractinsky, 2004). These dimensions describe
high-level attributes of web-page design as perceived by the
users. They are very comparable to dimensions that
describe the aesthetics of other environments (Kaplan,
1988; Nasar, 1988b). The dimensionsclassical and
expressiveare distinct (albeit correlated) aspects of web-
page aesthetics. Classical aesthetics refers to the orderli-
ness, or clarity of the design. Expressive aesthetics refers to
the creativity and the richness of the design. Thus, in this
experiment we examined whether the attractiveness of web
pages is related to their design as captures by the classical
and expressive aesthetics dimensions. In addition, the
experiment was conducted such that a comparison can be
made with Experiment 1 regarding the relative attractive-
ness of web pages. High correlations between attractiveness
ratings in different samples would increase the general-
izability of the studys ndings.
4.1. Method
4.1.1. Procedure
The procedure used for this experiment was similar to
Experiment 2 with two modications. In the second phase,
instead of once again rating the pages attractiveness (as in
Experiment 1) the participants rated the pages on 6 items
measuring expressive and classical aesthetics. The six items
were shortened versions of the original scales developed by
Lavie and Tractinsky (2004). We chose to use the short
version to reduce the effects of repetitiveness and fatigue on
users ratings. In both phases the web pages were presented
in a random order. Because the evaluations of classical and
expressive aesthetics require more reection than a single
attractiveness rating, the exposure time of web pages in the
second phase was not limited (as opposed to a 10 s limit in
Experiment 1).
4.1.2. Stimuli
Twenty-four web pages were selected from the 50 pages
used in Experiment 1. The pages were selected based on
their attractiveness ratings in that experiment. We selected
the eight most attractive pages, the eight least attractive
ARTICLE IN PRESS
1
1.5
2
2.5
1 3 5 6 7 9 10
Attractiveness Evaluation
L
a
t
e
n
c
y

(
s
e
c
.
)
Mean
Median
2 4 8
Fig. 4. Evaluation latencies as a function of attractiveness ratings.
Table 1
Pairwise (Tukeys HSD) post-hoc comparisons between transformed (ln)
latencies of web-page ratings
Rating category
a
Against category
a
3 2 1 0
4 (ratings of 1 and10) ns

3 (ratings of 2 and 9) ns ns

2 (ratings of 3 and 8) ns

1 (ratings of 4 and 7)

0 (ratings of 5 and 6)

po0.05.
a
Categories 04 represent the extremity of evaluations of web-page
attractiveness (0 middle of the scale; 4 ends of scale).
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1077
pages, and eight pages that were rated in the middle of the
pack. The 24 web pages are presented in the Appendix.
4.1.3. Sample
Fifty-three undergraduate engineering students (who did
not participate in Experiment 1) participated in this
experiment for class credit. The sample included 37 male
and 16 female students. The average age was 24.9 years.
4.1.4. Measures
In addition to an attractiveness measure obtained in
Phase 1, users evaluations of the web pages in Phase 2
were captured by 10-point rating scales. The rating scales
asked participants to mark their level of agreement with
statements regarding the pages design. The statements
included three expressive aesthetics statements (sophisti-
cated, creative and fascinating) and three classical aes-
thetics statements (clean, pleasant and aesthetic).
Cronbachs a reliabilities of these scales were computed
separately for each page. The reliabilities ranged from 0.70
to 0.92, with an average of 0.84 for both the classical and
the expressive scales over all 24 pages.
4.2. Results
The average attractiveness ratings after 0.5 s exposure
are presented below the thumbnail of each page in the
Appendix. The results of the attractiveness ratings were
compared to the results of the attractiveness ratings of the
same subset of pages gathered after the same exposure time
in Experiment 1. Fig. 5 plots the average attractiveness
ratings of the 24 pages in Experiments 1 and 2. Overall,
ratings were somewhat higher in this experiment
(mean 5.63, SD 1.35) compared to the average ratings
of the same 24 pages in Experiment 1 (mean 5.15,
SD 1. 56). Still, the average attractiveness evaluations of
the web pages in the two samples were highly correlated
(r
2
0.84).
The mean attractiveness ratings of the 3 groups of web
pages, after a .5 s exposure, were in accordance with
expectations (See Table 2). The attractiveness ratings of the
top-eight pages were higher than that of the middle group,
and the attractiveness of the bottom group was the lowest.
A repeated measures ANOVA with Attractiveness Group
as the dependent variable was highly signicant (F(1.70,
88.63) 250.993, po0.001, degrees of freedom are Green-
houseGeisser corrected). Tests of within-subjects con-
trasts between all groups were highly signicant as well
(po0.001).
Figure 6 presents three scores (attractiveness, classical
and expressive aesthetics) for each web page, averaged over
53 participants. The web pages are sorted in this gure
from left to right in a descending order of attractiveness
ratings (denoted by triangles). We split the data points in
Fig. 6 to two categories (rather than use the original three
categories) because such visualization supports better
insights than if we had preserved the original three-group
classication. (An analysis of 3 groups provides the same
results for the most attractive and for the least attractive
groups, with inconclusive results for the intermediate
group).
As can be seen clearly in Fig. 6, the three measures are
correlated: in general, the more attractive pages scored
higher on both aesthetic dimensions (r 0.86 with classical
aesthetics and r 0.95 with expressive aesthetics). These
results demonstrate the very strong relationships between
the very brief attractiveness ratings and the more elabo-
rated assessments of classical and expressive aesthetics.
Still, the most salient pattern observed in this gure
suggests that the least attractive pages (the right-hand side
of Fig. 6) are markedly low on expressive aesthetics
(denoted by rectangles). For example, the page with the
lowest expressive aesthetics score in the top half of
attractive pages was still 1.75 points above the page with
the highest expressive aesthetics score in the bottom half of
attractive pages. The differences in classical aesthetics
(denoted by diamonds) between the two groups were
similar, though less pronounced.
To assess the relations between attractiveness and the
two aesthetics dimensions we conducted two types of
ARTICLE IN PRESS
2
3
4
5
6
7
8
9
2 4 6 8
Experiment 1
E
x
p
e
r
i
m
e
n
t

2
3 5 7 9
Fig. 5. Average attractiveness ratings of 24 web pages after 0.5 s exposure
in Experiments 1 and 2.
Table 2
Mean ratings of attractiveness (in the rst phase, after 0.5 s exposure),
classical aesthetics and expressive aesthetics (in the second phase,
unlimited exposure time) of the top-8, middle-8 and bottom-8 web pages
Pages Attractiveness Classical
aesthetics
Expressive
aesthetics
Top-8 7.03 (0.88) 7.28 (0.97) 7.12 (1.01)
Middle-8 5.68 (0.87) 5.77 (0.93) 4.93 (0.80)
Bottom-8 4.19 (0.96) 4.80 (1.14) 2.88 (0.85)
Grand mean 5.63 (0.73) 5.95 (0.79) 4.98 (0.70)
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1078
analyses.
6
The rst analysis was designed to test whether
the evaluations of classical and expressive aesthetics
obtained in the second phase of this experiment were
associated with the three pre-dened attractiveness cate-
gories (8 web pages in each of the high, medium, and low
levels). We used a repeated measures multivariate analysis
of variance (MANOVA), with Attractiveness Group as a
three-level within-subject factor and expressive and classi-
cal aesthetics as the dependent variables. The multivariate
within-subjects test was highly signicant (Wilks
Lambda 0.088, F(4,206) 122.056, po0.001). The uni-
variate tests were highly signicant for both classical
(F(1.57, 81.83) 134.47, po0.001, Partial Z
2
0.721) and
expressive aesthetics (F(1.69, 87.80) 510.84, po0.001,
Partial Z
2
0.908) (GreenhouseGeisser corrected degrees
of freedom in both tests).
For both dependent variables, within-subjects contrasts
between the three groups of web pages were signicant
(po0.001). The differences were more pronounced on the
expressive aesthetics dimension (Partial Z
2
0.88 and 0.84
between the top group and the middle group and between
the middle group and the bottom group, respectively) than
on the classical aesthetics dimensions (Partial Z
2
0.77
and 0.45 for the same contrasts, respectively).
A second, complementary, analysis was designed to test
the association between the attractiveness ratings, which
were obtained in Phase 1, and the two aesthetic dimen-
sions, which were obtained in Phase 2. A linear regression
analysis was conducted for the entire sample of 24 web
pages and separately for the group of 12 most attractive
pages and for the group of 12 least attractive pages.
Attractiveness ratings were regressed on classical aesthetics
and expressive aesthetics scores. For the entire sample of
web pages, only classical aesthetics contributed signi-
cantly to the equation (standardized regression
coefcient 0.356, p 0.014). Considering only this pre-
dictor, the models adjusted R
2
was 0.20. Within the group
of the most attractive pages (the left-hand side in Fig. 6),
only expressive aesthetics contributing signicantly
(b 0.336, p 0.03). Considering only this predictor, the
models adjusted R
2
was 0.18. Within the group of the least
attractive pages (the right-hand side in Fig. 6), classical
aesthetics was the only signicant contributor (b 0.426,
p 0.003), with adjusted R
2
0.25. The results indicate
that overall, and among the least attractive pages, classical
aesthetics contributed more to explaining variations in
immediate attractiveness ratings. Expressive aesthetics
explained only the attractiveness variations within the
group of the more attractive pages.
5. Discussion
The major objective of this study was to demonstrate that
aesthetics impressions of a web site are formed after a very
brief exposure to the site and that that such impressions are
not transient. Experiment 1s ndings demonstrate that
users are able to form immediate and consistent evaluation
of the attractiveness of web pages. These evaluations were
very consistent across web pages. That is, the degree to
which web pages were regarded, on average, as attractive
after a very short exposure remained stable given a
considerably longer exposure, lending support to the
proposition that the relative attractiveness of web pages is
determined quickly (Lindgaard et al., 2006). These results
are in line with Zhang and Lis (2004b) ndings regarding
the immediate and continuous effect of the systems
affective quality on users cognition and usage patterns.
Past studies that have demonstrated the immediacy of
certain affective reactions (e.g. Duckworth et al., 2002) have
mostly been based on simple stimuli. This study and the
studies cited above indicate that such reactions also occur
for visual aspects of computer software, and especially of
web pages, which is considerably more complex.
Further support for the immediate aesthetic evaluation
proposition is provided by the response latency measure.
One of the studys major objectives was to nd convergence
between the explicit and the implicit measures of attrac-
tiveness. Such convergence was demonstrated for both
attractive and unattractive evaluations, in line with similar
ndings from different contexts (e.g. Bassili, 1996; Ostrom
and Gannon, 1996; Pham et al., 2001; Ritterfeld, 2002). In
addition to the specic contribution of this measure to the
research context, this study demonstrates the potential of
ARTICLE IN PRESS
1
2
3
4
5
6
7
8
9
10
Web Pages
R
a
t
i
n
g
Least Attractive Most Attractive
Attractiveness Classical Expressive
Fig. 6. Average ratings (N 53) of attractiveness (in the rst phase, after
0.5 s exposure), classical aesthetics and expressive aesthetics for each of 24
web pages, sorted by attractiveness rating. Vertical dashed bar separates
the top 12 and bottom 12 pages according to attractiveness level.
Horizontal dashed bar indicates the middle of the rating scales.
6
The analyses were conducted to test associations, not causality; hence,
the use of independent or dependent variables in these analyses do not
imply causes or effects.
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1079
response-latency as an easy-to-collect, unobtrusive measure
of preferences and attitudes in HCI research.
There are very few studies about individual differences in
reactions to aesthetics of IT products (see Hassenzahl,
2004). This research area is important if we assume that
users aesthetic preferences play a role in their evaluation of
web pages and in their subsequent interactions with web
sites. Thus, it is interesting to note the differences in the
participants average ratings of the set of 50 web pages.
Whereas some participants rated the entire set of web pages
as fairly unattractive (lower than 4 on a 110 scale), other
participants rated it much higher (close to 7 on that scale).
It is also important to note that while the relative
attractiveness of web pages (i.e. designed objects) remained
stable between the two phases of Experiment 1 when
averaged over evaluators (i.e. users), the evaluations within
individual users were less consistent. There are several
possible reasons for this nding: It may be that there are
individual differences in peoples ability to consistently rate
the attractiveness of objects, especially given such short
exposures as in the rst phase. In addition, people may
differ in their ability to distinguish nuances in design.
People who are better at that may have detected additional
information during the longer exposure times in Phase 2.
This may have led to lower consistency between the two
phases for these participants. Finally, the differences may
simply reect the statistical property of sample means to be
more consistent than single observations.
Thus, the results highlight the two sides of aesthetic
evaluations. On the one hand, evaluations over a sample of
users can provide a reliable and consistent measure of the
general attractiveness of web pages. At the same time,
individual users may differ in terms of their tastes and
evaluations of web pages. Hence, despite the benets of
designing for the average user, there is still room and need
for tailoring the visual design of web pages to various
users tastes. The proliferation of software skins reects the
demand for such aesthetic personalization of IT applica-
tions (Tractinsky and Zmiri, 2006).
Some promising work has already been done on the
question of what design characteristics affect evaluations of
web pages (Kim and Moon, 1998; Kim et al., 2003; Park
et al., 2005). In Experiment 2 we examined the association
of immediate rst impressions with two aesthetic dimen-
sions of web pages, classical aesthetics and expressive
aesthetics (see Fig. 6). In general, the results indicate that
positive immediate impressions were associated with high
levels of both aesthetic dimensions. However, unattractive
pages were associated mainly with very low levels of
expressive aesthetics. Perhaps the lack of expressiveness in
those web pages left a dull design that users could not
consider as attractive. In addition, the more attractive web
pages were associated to various degrees with the two
aesthetic dimensions. That is, the design dimensions
measured in this study do not offer golden rules for
designing attractive web pages. These results reect the
ndings that the relationships between various design
dimensions and perceptions of web sites were not stable
(Chen et al., 2002). Rather, they depend on the type of web
sites evaluated and the population from which users are
sampled. Thus, while some general conclusions can be
drawne.g. that less attractive pages are characterized by
very low levels of expressive aesthetics, or that highly
attractive pages reect high levels of both aesthetics
dimensionsthere are still many contingencies involved
in the creation of rst impressions. There is ample room for
future research to elaborate on this issue. For example,
cultural and individual differences, as well as the web
domain (Zhang et al., 2001) may be important determi-
nants of how users perceive the attractiveness of web sites.
While the ndings of this study relate to a presumably
minor issue of web site design, the implications may be far
reaching. Recent studies have argued that positive affect
improves decision making, trust and social interactions
(e.g. Isen (2001). Other studies found that decisions consist
of a mix of conscious and nonconscious processes (Bargh,
2004) and that the degree to which nonconscious processes
inuence choice processes is much greater than most
choice researchers believe (Fitzsimons et al., 2002).
Studies have also shown that ones affective states are
related to aesthetics of ones environment, be it in the
working place (Rafaeli and Vilnai-Yavetz, 2004), the home
or the neighbourhood (Nasar, 1988b), the store (Russell
and Pratt, 1980), or the web site (Kim et al., 2003; Zhang
and Li, 2004a). In line with Lindgaard et al. (2006), this
study supports a possible explanation for various studies
that found inuence of aesthetics on attitudes towards the
IT artifact (e.g. Schenkman and Jonsson, 2000; Tractinsky
et al., 2000; Lindgaard and Dudek, 2003; van der Heijden,
2003). Since aesthetic information is evaluated immedi-
ately, it is largely responsible for the users rst impres-
sions. Subsequently, new information tends to be processed
in a way that is biased towards those rst impressions
(Fitzsimons et al., 2002). And because the aesthetic
impressions are quite stable, there is probably a need for
signicant counter evidence on other system attributes (e.g.
usability, reliability, functionality) to alter the users rst
impressions.
The ndings do not imply that rst aesthetic impressions
are solely responsible for users attitudes towards web
pages, just as attitudes towards other humans are not
determined by aesthetic rst impressions alone (Eagly et
al., 1991). Many factors can potentially moderate the
relations between rst aesthetic impressions of an IT
artifact and the attitudinal or behavioural consequences
of the interaction (Tractinsky, 2004). Still, as the adage
goes, there is no second chance to make a rst impression.
6. Summary
The study fullled the four objectives set forth at the
beginning of this paper. First, we replicated Lindgaard
et al.s (2006) ndings using different web sites, different
users and a slightly different rating scale. Second, we
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1080
provided converging evidence from an independent,
implicit measure (response latency) to the premise that
users can consistently judge the attractiveness of web pages
even after very brief exposure. Third, we showed that while
average ratings of web-page attractiveness are highly
consistent, there is considerable variance in the degree to
which individual users are consistent in their evaluations.
Fourth, we found some associations between immediate
attractiveness ratings and more elaborated evaluations of
two aesthetic dimensions of web pagesclassical and
expressive. Overall, the ndings suggest that visual
aesthetics plays an important role in users evaluations of
web pages and of interactive systems in general.
Appendix
Thumbnails of 24 web pages used in Experiment 2,
sorted by average attractiveness ratings (shown below each
web page).
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1081
References
Aaker, D.A., Bagozzi, R.P., Carman, J.M., MacLachlan, J.M., 1980. On
using response latency to measure preference. Journal of Marketing
Research 17, 237244.
Bargh, J.A., 2004. Bypassing the will: towards demystifying the
nonconscious control of social behavior. In: Hassin, R.R., Uleman,
J.S., Bargh, J.A. (Eds.), The New Unconscious. Oxford University
Press, New York.
Bassili, J.N., 1996. The how and why of response latency measurement in
telephone surveys. In: Schwarz, N., Sudman, S. (Eds.), Answering
Questions. Jossey-Bass Publishers, San Francisco.
Bornstein, R.F., 1992. Subliminal mere exposure effects. In: Bornstein,
R.F., Pittman, T.S. (Eds.), Perception Without Awareness: Cognitive,
Clinical, and Social Perspectives. The Guilford Press, New York,
pp. 191210.
Chen, Q., Clifford, S.J., Wells, W.D., 2002. Attitude toward the Site II:
new information. Journal of Advertising Research, 3345.
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1082
Coates, D., 2003. Watches Tell More than Time. McGraw-Hill,
New York.
Cook, T.D., Campbell, D.T., 1979. Quasi-Experimentation. Houghton
Mifn Co., Boston.
Dashiell, J.F., 1937. Affective value distances as determinants of esthetic
judgment-times. American Journal of Psychology 50, 5767.
Dion, K., Berscheid, E., Walster, E., 1972. What is beautiful is good.
Journal of Personality and Social Psychology 24 (3), 285290.
Duckworth, K.L., Bargh, J.A., Garcia, M., Chaiken, S., 2002. The
automatic evaluation of novel stimuli. Psychological Science 13 (6),
513519.
Eagly, A.H., Ashmore, R.D., Makhijani, M.G., Longo, L.C., 1991. What
is beautiful is good, buty: a metaanalytic review of research on the
physical attractiveness stereotype. Psychological Bulletin 110 (1),
109128.
Ehrenberg, A.S.C., 1990. A hope for the future of statistics: MSOD.
American Statistician 44, 195196.
Fitzsimons, G.J., Hutchinson, J.W., Williams, P., Alba, J.W., Chartrand,
T.L., Huber, J., Kardes, F.R., Menon, G., Raghubir, P., Russo, J.E.,
Shiv, B., Tavassoli, N.T., 2002. Non-conscious inuences on consumer
choice. Marketing Letters 13 (3), 269279.
Greenwald, A.G., McGhee, D.E., Schwartz, J.L.K., 1998. Measuring
individual differences in implicit cognition: the implicit association
test. Journal of Personality and Social Psychology 74, 14641480.
Hamermesh, D., Biddle, J., 1994. Beauty and the labor market. American
Economic Review 84 (December), 11741194.
Hamermesh, D., Parker, A.M., 2005. Beauty in the classroom: professors
pulchritude and putative pedagogical productivity. Economics of
Education Review 24 (4), 369376.
Hassenzahl, M., 2004. The interplay of beauty, goodness and usability
in interactive products. HumanComputer Interaction 19 (4),
319349.
van der Heijden, H., 2003. Factors inuencing the usage of websites: the
case of a generic portal in the Netherlands. Information and
Management 40, 541549.
Herr, P.M., Page, C.M., 2004. Asymmetric association of liking and
disliking judgments: so whats not to like? Journal of Consumer
Research 30, 588601.
Hubbard, R., Armstrong, J.S., 1994. Replications and extensions in
marketingrarely published but quite contrary. International Journal
of Research in Marketing 11, 233248.
Isen, A.M., 2001. An inuence of positive affect on decision making in
complex situations: theoretical issues with practical implications.
Journal of Consumer Psychology 11 (2), 7585.
Kaplan, S., 1988. Perception and landscape: conceptions and misconcep-
tions. In: Nasar, J.L. (Ed.), Environmental Aesthetics: Theory,
Research, and Applications. Cambridge University Press, Cambridge.
Karvonen, K., 2000. The beauty of simplicity. In: Proceedings of the
ACM Conference on Universal Usability (CUU 2000), November
1617, 2000, Washington DC, USA.
Kim, J., Moon, J.Y., 1998. Designing towards emotional usability in
customer interfacestrustworthiness of cyber-banking system inter-
faces. Interacting with Computers 10 (1), 129.
Kim, J., Lee, J., Choi, D., 2003. Designing emotionally evocative
homepages: an empirical study of the quantitative relations between
design factors and emotional dimensions. International Journal of
HumanComputer Studies 59 (6), 899940.
Lavie, T., Tractinsky, N., 2004. Assessing dimensions of perceived visual
aesthetics of Web sites. International Journal of HumanComputer
Studies 60, 269298.
Lindgaard, G., Dudek, C., 2003. What is this evasive beast we call user
satisfaction? Interacting with Computers 15, 429452.
Lindgaard, G., Fernandes, G.J., Dudek, C., Brownet, J., 2006. Attention
web designers: you have 50 ms to make a good rst impression!.
Behaviour and Information Technology 25 (2), 115126.
MacLachlan, J., Czepiel, J., LaBarbera, P., 1979. Implementation of
response latency measures. Journal of Marketing Research 16,
573577.
Monk, 2004. The product as a xed effect fallacy. HumanComputer
Interaction 19 (4), 371375.
Nasar, J.L. (Ed.), 1988a. Environmental Aesthetics: Theory, Research,
and Applications. Cambridge University Press, Cambridge.
Nasar, J.L., 1988b. Perception and evaluation of residential street scenes.
In: Nasar, J.L. (Ed.), Environmental Aesthetics: Theory, Research,
and Applications. Cambridge University Press, Cambridge.
Norman, D.A., 2004a. Emotional Design: Why We Love (or Hate)
Everyday Things. Basic Books, New York.
Norman, D.A., 2004b. Introduction to the Special Issue on Beauty,
Goodness and Usability. HumanComputer Interaction 19 (4), 311318.
Ostrom, Gannon, 1996. Exemplar generation: assessing how respondents
give meaning to rating scales. In: Schwarz, N., Sudman, S. (Eds.),
Answering Questions. Jossey-Bass Publishers, San Francisco.
Park, S., Choi, D., Kim, J., 2004. Critical factors for the aesthetic delity
of web pages: empirical studies with professional web designers and
users. Interacting with Computers 16, 351376.
Park, S., Choi, D., Kim, J., 2005. Visualizing E-Brand Personality:
exploratory studies on visual attributes and E-Brand Personalities in
Korea. International Journal of HumanComputer Interactions 19 (1),
734.
Pham, M.T., Cohen, J.B., Pracejus, J.W., Hughes, G.D., 2001. Affect
monitoring and the primacy of feelings in judgment. Journal of
Consumer Research 28, 167188.
Porteous, J.D., 1996. Environmental Aesthetics: Ideas, Politics and
Planning. Routledge, London.
Postrel, V., 2002. The Substance of Style. Harper Collins, New York.
Rafaeli, A., Vilnai-Yavetz, I., 2004. Emotion as a connection of physical
artifacts and organizations. Organization Science 15 (6), 671686.
Ritterfeld, U., 2002. Social heuristics in interior design preferences.
Journal of Environmental Psychology 22, 369386.
Russell, J.A., Pratt, G., 1980. A description of the affective quality
attributed to environments. Journal of Personality and Social
Psychology 38 (2), 311322.
Schenkman, B.N., Jonsson, F.U., 2000. Aesthetics and preferences of web
pages. Behavior and Information Technology 19 (5), 367377.
Shipley, W.C., Norris, E.D., Roberts, M.L., 1946. The effect of changed
polarity of set on decision time of affective judgments. Journal of
Experimental Psychology 1, 237243.
Stocke , V., 2003. Measuring Information Accessibility and Predicting
Response-Effects: The Validity of Response-Certainties and Response-
Latencies, Sonderforschungsbereich 504, 03-33. University of Mannheim.
Tractinsky, N., 2004. Towards the Study of Aesthetics in Information
Technology. In: Proceedings of the 25th Annual International
Conference on Information Systems (ICIS). Washington, DC,
December 1215, pp. 771780.
Tractinsky, N., Zmiri, D., 2006. Exploring attributes of skins as potential
antecedents of emotion in HCI. In: Fishwick, P. (Ed.), Aesthetic
Computing. MIT Press, Cambridge, MA, pp. 405422.
Tractinsky, N., Shoval-Katz, A., Ikar, D., 2000. What is beautiful is
usable. Interacting with Computers 13, 127145.
Tyebjee, T.T., 1979. Response latency: a new measure for scaling brand
preference. Journal of Marketing Research 16, 96101.
Zajonc, R.B., 1980. Feeling and thinking: preferences need no inferences.
American Psychologist 35 (2), 151175.
Zhang, P., Li, N., 2004a. The importance of affective quality. Commu-
nications of the Association for Computing Machinery 48 (9),
105108.
Zhang, P., Li, N., 2004b. Love at rst sight or sustained effect? the role of
perceived affective quality on users cognitive reactions to IT. In:
Proceedings of the International Conference on Information Systems
(ICIS), Washington, DC.
Zhang, P., von Dran, G.M., 2000. Satisers and dissatisers: a two-factor
model for website design and evaluation. Journal of the American
Society for Information Science 51 (14), 12531268.
Zhang, P., von Dran, G., Blake, P., Pipithsuksunt, V., 2001. Important
design features in different website domains: an empirical study of user
perceptions. e-Service Journal 1 (1).
ARTICLE IN PRESS
N. Tractinsky et al. / Int. J. Human-Computer Studies 64 (2006) 10711083 1083