Journal of Personality and Social Psychology
1990, Vol. 59, No. 5,981-993
Copyright 1990 by the American Psychological Association, Inc.
0022-3514/90/S00.75
Gender Differences in Effects of Physical Attractiveness on Romantic
Attraction: A Comparison Across Five Research Paradigms
Alan Feingold
Yale University
Evolutionary and sociocultural theories of mate selection preferences contend that men place
greater value on physical attractiveness than do women. Thus, meta-analyses were conducted of
findings from 5 research paradigms that have examined the hypothesis: (a) questionnaire studies,
(b) analyses of lonely hearts advertisments, (c) studies that correlate attractiveness with opposite-sex
popularity, (d) studies that correlate attractiveness with liking by a dyadic interaction partner, and
(e) experiments that manipulate the attractiveness and similarity of an opposite-sex stranger. The
anticipated sex difference emerged in all five meta-analyses, although it was larger in research that
examined self-reports than in research that examined social behavior.
It is a popular belief that being physically attractive is of
greater importance for women than for men, particularly in
attracting the opposite sex. Research on the value placed on
physical appearance began in the 1930s and 1940s by sociolo-
gists who constructed mate selection questionnaires (Baber,
1936; Christensen, 1947; Hill, 1945; Strauss, 1946). Respon-
dents were asked to judge the importance of different charac-
teristics for their ideal mate, and more men than women re-
ported that "good looks" were necessary (see review by Powers,
1971).
Two correlational paradigms were subsequently developed to
address the same question. The attractiveness-popularity para-
digm examines the correlation between people’s physical attrac-
tiveness and their popularity with the opposite sex (e.g, Ber-
scheid, Dion, Walster, & Walster, 1971). The dyadic interaction
paradigm consists of the arrangement of encounters (e.g, blind
dates) between single men and women. Individuals’ physical
attractiveness is correlated with postinteraction reports of how
much they are liked by their partners (e.g, Walster, Aronson,
Abrahams, & Rottmann, 1966). Findings of sex differences ob-
tained in these paradigms were inconclusive for two reasons.
First, physical attractiveness is related to other characteristics,
including social skills (Feingold, 1990; Goldman & Lewis,
1977), and the effects of physical attractiveness are thus con-
founded with effects of other characteristics (Berscheid & Wal-
ster, 1974; Reis, Wheeler, Spiegel, Kernis, & Perri, 1982). Sec-
ond, sex differences were not consistent across studies. Ber-
scheid et al. (1971) and Krebs and Adinolfi (1975) found higher
correlations between attractiveness and dating popularity for
women than for men. However, Reis and his colleagues (Reis,
Nezlek, & Wheeler, 1980; Reis et al, 1982) found attractiveness
I want to thank David Buss, Judith Howard, and J. Richard Udry for
providing me with unpublished findings from their studies. I also want
to thank five anonymous reviewers for their comments on a draft of
this article.
Correspondence concerning this article should be addressed to Alan
Feingold, Department of Psychology, Yale University, P.O. Box 11A
Yale Station, New Haven, Connecticut 06520.
to be more strongly associated with the quantity of opposite-sex
interactions for male students, and Curran and Lippold (1975)
found similar correlations for both sexes. Inconsistencies of re-
sults were also found among studies of dyadic interactions
(compare, for example, Byrne, Ervin, & Lamberth, 1970, with
Curran, 1973). Consequently, a number of reviewers have con-
cluded that the evidence for the gender difference in the value
accorded to others’ attractiveness is inconclusive (Berscheid &
Walster, 1974; Huston & Levinger, 1978; Reis et al, 1980).
Another method of assessing effects of physical appearance
on attraction evolved from Byrne’s research with the bogus
stranger paradigm (see Byrne, 1971, for a review). In the attrac-
tiveness-similarity paradigm (e.g, Byrne, London, & Reeves,
1968), each subject evaluates a target who is either similar or
dissimilar to the subject (manipulated by bogus attitude scales)
and who is attractive or unattractive (manipulated by photo-
graphs appended to the target’s putative attitude responses).
The subject then evaluates the target on Byrne’s (1971) Interper-
sonal Judgment Scale (US). Two US ratings expressing liking
and the desire to work with the target in an experiment are
summed to compose the measure of attraction.
Unfortunately, most of the experiments that used this para-
digm were concerned with the main effects of attractiveness
rather than with sex-related interactions, and thus did not have
subjects of both sexes judge opposite-sex targets (Black, 1974;
Giarrusso, 1978; Insko et al, 1973; Kleck & Rubenstein, 1975;
Layton & Insko, 1974; Marcus, 1976; McWhirter, 1970). Other
studies combined ratings made of same- and opposite-sex tar-
gets in complex designs (Byrne et al, 1968; Rosenbaum, 1986),
and reported data in a manner that would require extensive
reanalyses before the effects of target attractiveness on oppo-
site-sex attraction would be revealed. Indeed, the only attractive-
ness^similarity experiment that focused on opposite-sex attrac-
tion was done by Stroebe, Insko, Thompson, and Layton
(1971), and that experiment has been widely cited as having
demonstrated that physical attractiveness is valued more by
men than by women. That conclusion was unwarranted be-
cause Stroebe et al. did not test that hypothesis. The researchers
used three levels (low, medium, and high) of target attractive-
981

982 ALAN FEINGOLD
ness. Thus, the hypothesized sex difference should have been
examined by crossing the linear trend component of the attrac-
tiveness factor with sex of subject. Instead, Stroebe et al. re-
ported the 2 # Attractiveness X Sex interaction that combined
linear and quadratic sources of variation. Thus, none of the
analyses of experiments using the attractiveness-similarity
paradigm has directly examined the sex difference in effect of
target attractiveness on opposite-sex attraction.
The most recently developed approach to the assessment of
effects of attractiveness on heterosexual attraction is the content
analysis of personal ads paradigm (e.g., Harrison & Saeed,
1977). The findings from such studies have shown that male
advertisers are more likely than female advertisers to be seeking
an attractive respondent (Hatfield & Sprecher, 1986).
In summary, the predicted sex difference emerged consis-
tently in only two of the five paradigms (questionnaires and
personal ads), both of which examine self-reported preference
for attractiveness and thus share methodological weaknesses. If
it is less socially desirable for women than for men to avow that
they value appearance, spurious gender differences might be
obtained. In addition, both types of self-reports convey infor-
mation about what people think they want in a romantic
partner rather than what is actually sought, and gender differ-
ences in the former may not reflect gender differences in the
latter. For example, both sexes may underestimate the value
they place on attractiveness (Berscheid & Gangestad, 1982), but
women may do so to a greater degree than do men, thus result-
ing in a larger sex difference in self-reported preference than in
social behavior.
The development of evolutionary theories of human social
behavior (Buss, 1989; Cunningham, 1986; Kenrick, Sadalla,
Groth, & Trost, 1990; Symons, 1979; Thiessen & Gregg, 1980;
Trivers, 1985) has afforded a strong theoretical framework for
sex differences in mate selection criteria. It has been noted that
a key biological difference between the sexes is that women
bear a limited number of offspring, whereas men can sire many
children (and over a longer portion of the life span). Thus, men
are said to maximize reproductive success by being sexually
responsive to larger numbers of partners, to whom they are
attracted by youth and beauty (cues to reproductive health).
Women are believed to judiciously choose a small number of
sexual partners who display social dominance (cues to resource
acquisition). Therefore, it has been suggested that there may be
a genetic basis for men’s mate choices to be determined mainly
by physical appearance (visual arousal) and for women’s prefer-
ences to be affected primarily by characteristics of men that
enhance the probability of the survival of their progeny.
The evolutionary perspective has resulted in a resurgence of
research mostly using questionnaires to examine gender
differences in mate preferences (e.g, Buss & Barnes, 1986; How-
ard, Blumstein, & Schwartz, 1987; Sadalla, Kenrick, & Ver-
shure, 1987; Townsend, 1989), which has confirmed earlier find-
ings that men report physical attractiveness to be more impor-
tant to them than do women. However, the issue of whether sex
differences in self-reports are generalizable to sex differences
in romantic behavior remains important. That question may
best be examined through meta-analysis (Hedges & Olkin,
1985; Rosenthal, 1984), which has become a popular method
for evaluating gender differences (e.g, Hyde & Linn, 1986).
Meta-analysis can suggest explanations for apparently con-
flicting results in the literature through the identification of
moderator variables. For example, seeming inconsistencies in
sex differences in attractiveness-popularity studies may be re-
solved by categorizing study findings by type of popularity.
Thus, this article uses meta-analysis to integrate sex differences
in importance of attractiveness for each of the five paradigms
described and to compare sex differences from self-report
paradigms with sex differences from behaviorally oriented par-
adigms.
Retrieval of Studies
Dissertation Abstracts International, ERIC, the Inventory of
Marriage and Family Literature, Masters Abstracts, PsychLIT,
and Sociofile were searched to locate post-1960 studies that
used at least one of the five paradigms of interest with Ameri-
can or Canadian subjects. Reference lists of all retrieved studies
and of reviews of the attractiveness literature (e.g. Bull & Rum-
sey, 1988; Cash, 1981) were also examined.
Analysis 1:
Mate Selection Questionnaire Paradigm
Method
The literature search retrieved 23 studies (examining 28 indepen-
dent samples) that had administered a nonnativefy scaled* mate selec-
tion questionnaire to both sexes and reported results that allowed the
effect size for the gender difference in preference for attractiveness to
be calculated using procedures described by Hedges and Becker (1986)
and Rosenthal (1984).2 When means and standard deviations were
reported (or could be obtained from reported frequency distributions)
by sex, the effect size d(Cohen, 1977) was calculated by subtracting the
female mean from the male mean and dividing the difference by the
pooled within-sex standard deviation. Thus, positive values oft/ indi-
cated that male respondents rated attractiveness (usually termed good
looks in the questionnaires) as being more important to them than did
female respondents. For some studies, d was calculated from reported;
or F ratios. Where information was inadequate to compute the exact
effect size, standard estimation procedures were used when feasible. If,
1
Two types of scaling procedures have been used in the construc-
tion of mate selection questionnaires: the normative method, in which
each characteristic (e.g., physical attractiveness) is rated for importance
independently of the others (e.g. Buss & Barnes, 1986, Study 1; Howard
et al, 1987), and the ipsalive method, in which different attributes are
ranked in varied ways, resulting in the assessment of the relative impor-
tance of different attributes among a specific set of characteristics (e.g.
Buss & Barnes, 1986, Study 2; Daniel, O’Brien, McCabe, & Quinter,
1985; Hendel, 1978; Laner, 1977). Because the meta-analysis is con-
cerned with gender differences in absolute value placed on attractive-
ness, studies that used ipsative instruments were excluded.
2
The effect sizes for the gender differences from three included
studies (Buss & Barnes, 1986, Study 1; Feingold, 1981; Howard et al,
1987) could not be determined from information in the published
work and were calculated from unpublished raw data provided by the
authors.