Beauty

Abstract

Beauty, which the Medical Subject Headings classify as a kind of esthetics, is the set of attributes of a person or thing that elicit pleasurable feelings, and cognitive psychology studies it as a measurable response of the perceiver rather than a property residing in the object alone. Research on faces has made beauty one of the most quantified topics in the field: observers agree substantially about who is attractive, and that agreement tracks measurable features. Averageness, symmetry, and sexual dimorphism each raise rated attractiveness, and evolutionary accounts read these cues as signals of mate quality. Neuroimaging locates the response in the brain's reward circuitry, while twin and individual-difference studies show a large private component riding atop the shared consensus. This article surveys agreement, averageness, symmetry, reward, and the balance of shared and private taste.

Keywords: beauty, facial attractiveness, averageness, symmetry, evolutionary psychology

Beauty, in the sense cognitive psychology studies, is not a philosophical absolute but an empirical regularity: the characteristics of faces, bodies, and objects that reliably elicit pleasure across observers (Langlois et al., 2000). The Medical Subject Headings define it as the characteristics or attributes of persons or things that elicit pleasurable feelings, and file it beneath esthetics, the empirical science of beauty from which this topic inherits its methods. Where philosophical aesthetics asks what beauty is, the psychology of beauty asks which features predict it and what the mind does in response, and it has answered most fully for the human face (Rhodes, 2006; Little et al., 2011).

What Beauty Is

Beauty denotes the attributes of a person or object that elicit pleasure in an observer, together with the evaluative response those attributes evoke. The Medical Subject Headings place it beneath esthetics, reflecting its descent from the empirical science of beauty that Gustav Fechner founded with his programme of 'aesthetics from below', the measurement of ordinary preferences in place of philosophical speculation (Fechner, 1876). For cognitive psychology the useful reading is the empirical one: a judgment of beauty is an act of evaluative cognition computed from the interaction of a stimulus with a perceiver's perceptual apparatus, knowledge, and state (Langlois et al., 2000; Little et al., 2011).

What makes beauty tractable, and the reason most of its science is a science of faces, is that it can be measured with unusual precision. Observers rate attractiveness quickly, reliably, and with high agreement, so the trait behaves as a stable dependent variable that can be predicted from measurable properties of the stimulus (Langlois et al., 2000). The sections that follow trace the main predictors the field has found, averageness, symmetry, and sexual dimorphism, the evolutionary logic offered to explain them, the reward response they evoke, and the private taste that the consensus leaves unexplained.

'What Is Beautiful Is Good': Beauty as a Social Signal

Before turning to what makes a face beautiful, it is worth marking what beauty does, because its consequences are what make the topic more than cosmetic. Dion, Berscheid, and Walster demonstrated the physical-attractiveness stereotype: observers attribute to attractive people a range of unrelated positive traits, assuming they are kinder, more sociable, more competent, and likely to lead happier lives (Dion et al., 1972). The phrase they coined, 'what is beautiful is good', names a halo in which a single salient positive quality colours judgment of every other.

The stereotype is consequential rather than merely decorative. Attractiveness shapes hiring, sentencing, and election outcomes, and Langlois and colleagues' meta-analysis confirmed that it is both widely held and partly accurate in its correlates, since attractive people do receive more favourable treatment from infancy onward (Langlois et al., 2000). That beauty functions as a social signal of this power is the practical reason its determinants are worth measuring, and it is to those determinants that the remaining sections turn.

Averageness and Attractiveness

The single most robust finding in the science of facial beauty is also the most counterintuitive: average faces are attractive. Langlois and Roggman digitally averaged photographs of many faces and found that the composites were rated more attractive than almost all of the individual faces that composed them, and that attractiveness rose as more faces entered the average (Langlois & Roggman, 1990). Averageness here is not blandness but mathematical centrality: averaging cancels idiosyncratic deviations and pulls the composite toward the population prototype.

Why should the prototype be beautiful? One answer is cognitive: prototypical stimuli are processed fluently, and fluent processing is hedonically marked, so proximity to the learned average is experienced as pleasure (Reber et al., 2004). The same fluency logic underlies the mere-exposure effect, Zajonc's finding that repeated exposure to a stimulus raises liking for it, since a face near the population average is in effect one every observer has been exposed to most often (Zajonc, 1968). A second answer is evolutionary: averageness may signal genetic heterozygosity and developmental stability, making it a cue to mate quality. The two explanations are not exclusive, and both predict the same ordering of preferences. Averageness is not the whole story, since symmetry and specific dimorphic features add beyond it, but it is the firmest single regularity the field has (Rhodes, 2006; Jones & Jaeger, 2019). The model below builds composite faces from a chosen number of exemplars, showing how averaging pulls the result toward the prototype and smooths away individuating deviation.

Symmetry and Sexual Dimorphism

Two further regularities contribute beyond averageness. The first is bilateral symmetry. Faces whose left and right halves match more closely are rated more attractive, and because developmental perturbations, disease, parasites, and mutations, tend to produce fluctuating asymmetry, symmetry has been read as a visible record of the ability to develop stably despite such insults (Rhodes, 2006; Thornhill & Gangestad, 1999). The effect is real but modest, and it is partly confounded with averageness, since symmetric faces are also closer to the prototype.

The second is sexual dimorphism, the degree to which a face carries sex-typical features. Perrett and colleagues used computer-graphic caricaturing to exaggerate or reduce the shape differences between average male and female faces, and found that feminising a female face raised its attractiveness beyond the average, a result that complicated the simple averageness account (Perrett et al., 1998; Perrett et al., 1994). Masculinity in male faces produces a more divided response, sometimes preferred, sometimes penalised as a cue to lower warmth, and preferences shift with context and with the perceiver's own state (Little et al., 2011). A recent large replication confirmed symmetry, averageness, and femininity as independent contributors to female facial attractiveness while finding their individual effects smaller than early studies implied (Jones & Jaeger, 2019). The model below lets symmetry vary continuously on an abstract face and plots the predicted rise in rated attractiveness.

The Evolutionary Account

Why these particular features, and why so consistently? The dominant theoretical frame is evolutionary. On this account the preferences are adaptations for mate choice: averageness, symmetry, and sex-typical features are honest signals of heritable quality, because only a developmentally robust, healthy, parasite-resistant organism can afford to grow them, so a preference for them would have raised reproductive success (Thornhill & Gangestad, 1999; Little et al., 2011). The framework ties the disparate cues to a single logic and generates testable predictions, for instance that preferences should strengthen when the mating stakes or disease pressures are higher.

The account is influential but not settled. The link between the attractive cues and actual health or fertility is weaker and less consistent than the 'good genes' logic requires, and several predicted correlations have failed to replicate at scale (Rhodes, 2006). Cunningham's early multivariate work already showed that attractiveness is not a single dimension but a weighted blend of neonate, maturity, and expressive features, some of which fit the signalling story awkwardly (Cunningham, 1986). The evolutionary frame remains the field's most productive source of hypotheses while its strongest claims stay contested, a tension the Discussion returns to. The honest-signal logic that underwrites the whole account is a single causal chain, summarised in Figure 1.

Figure 1

The Reward Value of Beauty

Whatever its origin, beauty is processed by the brain as a reward. Aharon and colleagues combined functional imaging with a behavioural keypress task and found that beautiful faces activate reward circuitry, including the nucleus accumbens, and that heterosexual men would work, pressing a key repeatedly, to prolong their view of attractive female faces, treating the images like other rewarding stimuli (Aharon et al., 2001). Beauty, in other words, is not merely judged; it is wanted, and the wanting has a measurable neural and behavioural signature.

This reward framing connects the psychology of beauty to the wider neuroscience of valuation. Said and Todorov built a data-driven statistical model of facial attractiveness, representing faces in a multidimensional feature space and showing that attractiveness could be predicted from a face's position in that space, with an optimum near, but not identical to, the average (Said & Todorov, 2011). Their model makes the reward value of a face a computable function of its measurable features, bringing beauty within reach of the same quantitative tools used for other rewards (Chatterjee & Vartanian, 2016). The reward signal is not confined to the valuation core: Vessel and colleagues found that intensely moving aesthetic experiences also engage the default-mode network, a system usually silent during external stimulation, and that this response generalises across visual domains rather than being special to faces (Vessel et al., 2019).

Shared and Private Taste

The emphasis on consensus can mislead, because agreement about beauty, though real, is far from total. Hönekopp partitioned the variance in facial-attractiveness ratings into a shared component, the consensus every observer contributes to, and a private component, the part idiosyncratic to the individual perceiver, and found that private taste accounted for roughly as much variance as shared taste (Hönekopp, 2006). 'Beauty is in the eye of the beholder' turns out to be about half true.

Where does private taste come from? Germine and colleagues put the question to a large twin sample and found that individual aesthetic preferences for faces are shaped mostly by personal, non-shared environment rather than by genes or family, so the idiosyncratic half of beauty is largely learned from an individual's own experience of particular faces (Germine et al., 2015). The two findings fit together: a species-typical consensus, plausibly shaped by evolution and by exposure to a common population of faces, is overlaid by a private preference each person acquires from their unique visual history. The model below lets the reliability of a single rater vary and shows, by the Spearman-Brown relation, how averaging many raters recovers the shared consensus while private taste averages away.

Worked Example

The shared-versus-private split has a direct, quantitative consequence for how beauty is measured, which the third model above renders. Suppose a single observer's attractiveness rating of a face correlates only moderately with the underlying consensus, so that a single rater's reliability is 0.50, meaning half of one person's rating variance reflects shared taste and half is private idiosyncrasy. This is close to what Hönekopp reported (Hönekopp, 2006).

How reliable is the average rating of several observers? The Spearman-Brown relation gives the reliability of a mean of n raters as n times the single-rater reliability, divided by one plus n minus one times that reliability. With a single-rater reliability of 0.50, a panel of 4 reaches 0.80, a panel of 10 reaches 0.909, a panel of 20 reaches 0.952, and a panel of 40 reaches 0.976. Averaging cancels the private component because it is uncorrelated across observers, while the shared component survives, so a modest panel measures the consensus with high reliability.

Two lessons follow, and both match the literature. First, the high inter-rater agreement that Langlois and colleagues documented is partly a property of aggregation: individual agreement is middling, but averaged ratings are highly reliable, which is why composite judgments look so consensual (Langlois et al., 2000). Second, the gain diminishes, the jump from 1 to 4 raters buys 0.30 of reliability, but from 20 to 40 only a further 0.024, so a panel of ten to twenty observers already captures most of the recoverable consensus, which is why attractiveness studies rarely need large rating panels.

Discussion

The determinants surveyed here are complementary rather than competing. Averageness, symmetry, and sexual dimorphism each pick out a genuine, partly independent contributor to rated attractiveness, and the reward and statistical-model findings describe what the brain does with the resulting signal (Rhodes, 2006; Said & Todorov, 2011). A useful way to hold them together is to note that each identifies a different level of the same response: averageness and symmetry are properties of the stimulus, the reward signal is the perceiver's neural response, and the shared/private split describes how that response is distributed across a population of perceivers.

The through-line from Dion to the statistical face models is the conviction that beauty is a response of the perceiver that can be measured, predicted, and explained in terms of ordinary cognition and its evolutionary history, not a property residing in the object alone (Langlois et al., 2000; Hönekopp, 2006). That stance remains the field's unifying commitment, and its sharpest open question is how the species-typical consensus and the privately learned half are weighted and integrated for a given observer viewing a given face.

Current Directions

The most active front is computational and individual. Where early work sought the universal features of beauty, recent research models the individual perceiver: Germine and colleagues' twin study established that the private half of face preference is environmentally acquired, and subsequent work has pursued how a person's visual diet sculpts their idiosyncratic taste (Germine et al., 2015). The statistical-model tradition, meanwhile, has moved from predicting consensus attractiveness to fitting each observer's own preference function in a measured face space, and Iigaya and colleagues have shown that aesthetic preference can be predicted from a mixture of low- and high-level visual features extracted by a deep network, decomposing a judgment once treated as holistic into computable parts (Said & Todorov, 2011; Iigaya et al., 2021).

A second strand revisits the biological claims with better samples. Large, pre-registered replications such as Jones and Jaeger's have confirmed that symmetry, averageness, and sexual dimorphism contribute independently to female facial attractiveness, while estimating their effects as smaller and more variable than the foundational studies implied, a quieter but healthier evolutionary aesthetics (Jones & Jaeger, 2019). A third questions how automatic the response is at all: Brielmann and Pelli found that the experience of intense beauty depends on the availability of cognitive resources, so that a distracting secondary task suppresses it, evidence that beauty 'requires thought' rather than arising pre-attentively (Brielmann & Pelli, 2017). Running beneath these strands is a broader reframing from neuroaesthetics that treats beauty not as a special faculty but as one expression of the brain's general machinery for hedonic valuation, a programme some argue should dissolve 'aesthetics' as a distinct object of study altogether (Chatterjee & Vartanian, 2016; Skov & Nadal, 2020).

Key Researchers

Lisa M. DeBruine

(contemporary). With Jones and Little, has driven the modern evolutionary study of facial attractiveness, using computer-graphic face manipulation to test how preferences for symmetry and sexual dimorphism shift with context and perceiver state. ORCID

Nancy L. Etcoff

(contemporary). A cognitive neuroscientist who brought facial beauty to a wide audience and co-authored the reward-value imaging work showing that beautiful faces engage the brain's reward circuitry. ORCID

Laura T. Germine

(contemporary). Led the twin study establishing that individual aesthetic preferences for faces are shaped mostly by personal, non-shared environment, locating the private half of beauty in individual experience. ORCID

Judith H. Langlois

(contemporary). Discovered the averageness effect with Roggman and led the field-defining meta-analysis establishing that attractiveness judgments show substantial agreement within and across cultures and carry real social consequences. Faculty page

David I. Perrett

(contemporary). Pioneered computer-graphic facial-shape manipulation, showing with May and Yoshikawa that attractiveness can exceed the average and that sexual dimorphism shapes judgments of female and male faces. ORCID

Gillian Rhodes

(contemporary). Authored the authoritative review of the evolutionary psychology of facial beauty, synthesising the evidence on averageness, symmetry, and sexual dimorphism and weighing the good-genes account against its critics. Faculty page

Randy Thornhill

(contemporary). An evolutionary biologist who, with Gangestad, framed facial attractiveness as a signal of developmental stability, making fluctuating asymmetry a central variable in the evolutionary account of beauty. ORCID

Alexander Todorov

(contemporary). Built data-driven statistical models of face evaluation, with Said representing attractiveness as a computable function of a face's position in a measured feature space. ORCID

Glossary

Averageness Effect.

The finding that composite faces formed by averaging many individuals are rated more attractive than most of their constituents, because averaging yields a near-prototypical form.

Beauty.

The characteristics or attributes of a person or thing that elicit pleasurable feelings in an observer, studied empirically as a measurable response of the perceiver.

Collative Consensus.

The shared component of attractiveness judgments on which observers agree, as distinct from the private taste idiosyncratic to each perceiver.

Developmental Stability.

The capacity of an organism to develop its intended form despite genetic and environmental perturbation; low fluctuating asymmetry is read as its visible sign.

Face Space.

A multidimensional representation in which each face is a point defined by measured features, used to model attractiveness as a function of a face's position.

Fluctuating Asymmetry.

Small, random left-right departures from bilateral symmetry taken to index the developmental perturbations an organism has withstood.

Good-Genes Hypothesis.

The evolutionary claim that attractive cues are honest signals of heritable quality, so a preference for them raised ancestral reproductive success.

Halo Effect.

The tendency for one salient positive quality, such as physical attractiveness, to bias judgment of a person's unrelated traits, captured in 'what is beautiful is good'.

Physical-Attractiveness Stereotype.

The belief that attractive people also possess desirable social and intellectual traits, demonstrated by Dion, Berscheid, and Walster.

Private Taste.

The component of a person's attractiveness judgments unique to that individual, uncorrelated across observers and shaped mostly by personal experience.

Processing Fluency.

The subjective ease with which a stimulus is perceived and understood, experienced as positive affect and offered as one reason prototypical faces are liked.

Prototypicality.

The degree to which a face resembles the central tendency of its category; prototypical faces are processed fluently and tend to be preferred.

Reward Value.

The capacity of a stimulus to engage the brain's reward circuitry and motivate approach; beautiful faces carry reward value and elicit effort to view them.

Sexual Dimorphism.

The degree to which a face carries sex-typical shape; femininity raises female facial attractiveness, while masculinity preference in male faces is context-dependent.

Spearman-Brown Relation.

The formula giving the reliability of an averaged rating from the reliability of a single rater, explaining why a modest panel measures the beauty consensus reliably.

Symmetry.

The correspondence of a face's left and right halves; greater bilateral symmetry modestly raises rated attractiveness and is read as a cue to developmental stability.

Frequently Asked Questions

What is beauty in psychology?

It is the set of attributes of a person or thing that elicit pleasurable feelings, studied empirically as a measurable response of the perceiver rather than as a property of the object. Cognitive psychology treats attractiveness as a dependent variable to be predicted from features of the stimulus and the perceiver (Langlois et al., 2000).

Do people actually agree about who is beautiful?

Substantially, yes. Langlois and colleagues' meta-analysis found high agreement about facial attractiveness both within and across cultures, though a large private component remains: Hönekopp showed that individual, idiosyncratic taste accounts for roughly half the variance in ratings (Langlois et al., 2000; Hönekopp, 2006).

Why are average faces attractive?

Langlois and Roggman found that faces digitally averaged from many individuals are rated more attractive than their constituents. Averaging cancels idiosyncratic deviations and yields a near-prototypical face, which is processed fluently and may also signal genetic quality (Langlois & Roggman, 1990; Reber et al., 2004).

Does symmetry make a face more attractive?

Yes, modestly. More symmetric faces are rated more attractive, and because developmental disturbances produce asymmetry, symmetry has been interpreted as a cue to developmental stability, though the effect is small and partly confounded with averageness (Rhodes, 2006; Jones & Jaeger, 2019).

What is the 'what is beautiful is good' effect?

It is the physical-attractiveness stereotype: Dion, Berscheid, and Walster showed that observers attribute unrelated positive traits, such as kindness and competence, to attractive people, so beauty casts a halo over judgment of other qualities (Dion et al., 1972).

Is beauty rewarding to the brain?

Yes. Aharon and colleagues found that beautiful faces activate reward circuitry and that people will expend effort, pressing a key, to prolong their view of attractive faces, so beauty behaves like other rewarding stimuli (Aharon et al., 2001).

Is a preference for beauty innate or learned?

Both. A species-typical consensus exists, but Germine and colleagues' twin study found that individual face preferences are shaped mostly by personal, non-shared environment rather than genes, so the idiosyncratic half of taste is largely learned from experience (Germine et al., 2015).

Does the evolutionary account of beauty hold up?

Partly. It is the field's most productive source of hypotheses, tying averageness, symmetry, and dimorphism to mate-quality signalling, but the predicted links between attractive cues and actual health are weaker and less consistent than the strong version requires (Thornhill & Gangestad, 1999; Rhodes, 2006).

References