What Makes a Face Beautiful and How Much of It Is Biology?

One thing that reframed how I think about facial beauty is this: babies who are days old prefer to look at the same faces that adults rate as attractive.

No cultural exposure. No magazines, no Instagram, no parents telling them who's pretty. Just a tiny human who can barely see, spending more time looking at faces that happen to be the ones adults independently rate as beautiful.

That finding, combined with cross-cultural studies showing isolated tribes agree with Americans and Russians on attractive features, sets up the question this post is about: is "beauty is socially constructed" even the right framing? The pop-science answer is yes — Instagram and magazines invented modern beauty standards. The strong-biology answer is also no — beauty is hardwired by sexual selection. Both are wrong in opposite ways, and the interesting question is what's left when you take the strongest version of each.

My claim is: culture shapes the surface details, but underneath there's a biological signal consistent enough that newborns and isolated foragers detect it without media exposure — and that signal itself may be partly arbitrary, the product of aesthetic runaway rather than honest fitness signaling. That last part is the move most pop-science accounts miss.

Where do beauty standards come from?

When you look at a face and make an attractiveness judgment — which happens in about 100 milliseconds, before conscious thought kicks in — what are you actually evaluating? Where did that evaluation come from?

Two research strategies help disentangle biology from culture:

1. Cross-cultural studies — if isolated cultures that have never seen Western media still find the same features attractive, that's evidence for biology.
2. Infant studies — if babies who haven't been "contaminated" by culture show the same preferences as adults, that's evidence for biology.

Both have famous counterarguments — the tribes weren't actually isolated, the babies had hours to learn — and the rest of this post takes those counterarguments seriously rather than waving them away.

Babies prefer pretty faces before they know what pretty means

Judith Langlois's research established this clearly. You show babies two faces side by side — one that adults have rated as attractive, one as less attractive — and measure which one the baby looks at longer.

The results are consistent. Babies look longer at faces adults find attractive. This works across biological males and females and across different races and ethnicities. It is reliable as early as 3 months and possibly as early as 3 days. One-year-olds play almost twice as long with dolls that have attractive faces. Even more striking: infants prefer averaged composite faces — faces created by blending many faces together — over individual faces, with no experience required.

As Nancy Etcoff writes in Survival of the Prettiest: "Even babies, who have no sexual urges at all, are suckers for a pretty face and prefer to look at one from the very start."

The obvious counterargument: newborns have hours of caregiver exposure before any of these studies test them. Bushnell, Sai & Mullin (1989) showed that neonates can recognize their mother's face within roughly 12 hours of birth, even with olfactory cues controlled for. So "infants prefer attractive faces before any learning" is overstated. But what the studies do still establish is something narrower and harder to explain away: the preference is not idiosyncratic to the mother's specific features. Babies prefer faces of strangers, of different races, of different sexes — including faces they cannot have been exposed to in their first hours. The preference is for the population's central tendency, not for what they happened to see. That points at architecture, not exposure.

A related finding sharpens the point. Pascalis, de Haan & Nelson (2002) showed that 6-month-olds can discriminate individual monkey faces, but 9-month-olds and adults cannot — face processing tunes itself to whatever face population you grow up surrounded by, fast. So infant face perception is shaped by experience. But the shape it's tuned toward — the preference for averaged, symmetric, prototypical-of-the-environment faces — appears to be the default that experience refines, not the thing experience constructs.

Isolated tribes agree with Americans on beauty

The strongest cross-cultural evidence comes from two converging lines of work.

Cunningham and colleagues (1995) had recently arrived Asian and Hispanic students plus White Americans rate the attractiveness of Asian, Hispanic, Black, and White faces. The mean cross-group correlation was r = .93. A follow-up with Taiwanese raters produced r = .91. Self-reported exposure to Western media did not predict ratings. That's a near-ceiling correlation between people who grew up in entirely different visual environments.

Jones & Hill (1993) pushed further by studying Aché natives from Paraguay and Hiwi natives from Venezuela — populations the researchers framed as "isolated from the rest of the world and from each other." Across Aché, Hiwi, Americans, and Russians, everyone preferred women with large eyes and delicate jaws — features associated with youthfulness. The Aché and Hiwi agreed more with each other than with Americans and Russians, but they did agree with the Westerners.

The serious counterargument: the Aché and Hiwi were not truly isolated by 1993. Hill & Hurtado's Aché Life History, the canonical ethnography by one of the original researchers, documents decades of intermittent missionary and outside contact before the attractiveness study was run. And Tovée et al.'s work on Zulu and South African body preferences shows that BMI preferences shift toward Western values within roughly 18 months of migration. So "isolated cultures agree with us" is partly an artifact of how "isolated" we're willing to call populations who had real, sustained contact.

That doesn't sink the cross-cultural argument — it refines it. The Aché–Hiwi agreement with each other (the two groups that resembled each other physically but had no contact between them) is still meaningful evidence for what Jones & Hill called a "mere exposure" effect on facial preferences: you grow attached to the kind of face you grow up around. What survives the isolation critique is agreement on youthfulness markers (large eyes, delicate jaws, smooth skin), which appear across every population tested and don't track exposure. The universal isn't "the same supermodel"; it's "a small set of structural cues that index reproductive age."

Culture exploits biological predispositions

If biology provides a seed preference, culture grabs it and runs. The technical name for this is the supernormal stimulus, a concept from Tinbergen's 1948 paper on social releasers: once an animal evolves a response to a particular cue, you can often produce an even stronger response by exaggerating that cue beyond anything that exists in nature. Tinbergen's herring-gull chicks pecked harder at giant fake red dots than at real beaks. Deirdre Barrett's Supernormal Stimuli extends this logic to humans: most modern beauty culture is supernormal stimuli built on top of evolved preferences.

• Wider hips — a waist-to-hip ratio that genuinely signals fertility gets amplified by the Kardashians into surgically exaggerated proportions.
• Muscle mass — a fitness level that honestly signals health gets pushed into bodybuilding extremes far beyond functional strength.
• Large eyes — a youthfulness signal gets exploited by Disney animation (those enormous eyes would be terrifying in real life), mascara, and eyelid surgery.
• Smooth skin / less facial hair — sexual dimorphism that distinguishes biological females gets turned into a cultural expectation of complete hair removal.

Taken together, these examples make a specific claim about how culture builds on biology: it doesn't invent the preference from scratch, but it has no obligation to stop where biology would. The original signal was honest — wider hips actually correlate with reproductive capacity — but the cultural amplification is pure caricature.

Beauty standards that look purely cultural often turn out to be the same biological signal flipped by environment. A TED talk by Delali Bright recounts being bullied in West Africa for being thin and then arriving in America to find she was suddenly considered beautiful. Same body, opposite valence. The environmental security hypothesis (more on this later) explains why: in food-insecure environments, body fat signals survival; in food-secure environments, it signals lack of self-control. The signal flips because the ecological meaning of the cue flips. The underlying machinery — using body shape to infer "is this person doing well?" — doesn't.

The uncomfortable history of measuring faces

The science of facial attractiveness has a dark origin story. The first attempts to systematically measure faces were deeply intertwined with racist ideologies.

Petrus Camper measured the "facial angle" of Greek statues at 100° and ranked races by how close they came to this ideal. Western Europeans were closest. This was linked to intelligence.

Johann Caspar Lavater claimed specific features predicted character: "A weak chin is seldom found in discreet, well-disposed, firm men." The captain of the Beagle, a Lavater fan, almost rejected Darwin for the voyage because of his nose — Darwin's own autobiography records FitzRoy's doubt that "anyone with my nose could possess sufficient energy and determination."

Katherine Blackford linked hair color to personality: "The normal blonde is positive, dynamic, driving, aggressive... while the normal brunette is negative, static, conservative."

Francis Galton wanted to identify criminals by their faces. He overlaid photographs of convicted criminals onto composites, hoping to reveal the "criminal prototype." Instead, in his 1878 Nature paper on composite portraits, he accidentally discovered that the more faces you average together, the more attractive the composite becomes. The composite criminal was more attractive than any individual criminal.

The pattern here is more interesting than the racism. Every one of these men set out to confirm an ideology they already held and used measurement to launder it. Galton's accident — finding beauty where he was looking for criminality — is one of the few times in the history of physiognomy that the data refused to cooperate with the prejudice driving the project. That single accidental discovery, that averaged faces are attractive, turned out to be one of the most robust findings in the field.

Why averaged faces are attractive (but not the most attractive)

Galton's composite finding has been replicated extensively with modern methods. Langlois & Roggman (1990) digitized faces and showed that 16-face and 32-face composites are consistently rated more attractive than nearly all the individual faces that compose them.

But averaged faces are not the peak of attractiveness. They rank about 3–4 on a 5-point scale — consistently attractive, but not extraordinary. The most attractive faces — supermodel faces — actually deviate from the average in predictable ways: larger eyes, thinner jaws, smaller mouth-to-chin distance.

Supermodels have a specific combination: features of youthfulness (large eyes, small chin, smooth skin — typical of girls under 10) sprinkled with markers of sexual maturity (high cheekbones, which emerge at puberty). It's a biologically impossible combination — which is exactly what makes it compelling. The face signals youth and reproductive maturity simultaneously, which no real person can fully be. Supernormal stimulus, again.

Symmetry: attractive, but don't overdo it

Symmetry is the second biological standard, separate from averageness. Even when you control for the smoothing effect of averaging (which incidentally increases symmetry), symmetry independently predicts attractiveness. Studies with identical twins — where you compare the more symmetric twin to the less symmetric one — confirm this.

But there's a catch: perfect symmetry is creepy. If you take a face and mirror it exactly (creating a "chimera"), the result looks uncanny rather than beautiful. This connects to the uncanny valley — Masahiro Mori's 1970 observation that things approaching human likeness become appealing up to a point, then suddenly revolting when they get close but not quite.

Mori's curve was a hypothesis without quantitative data for forty years. Mathur & Reichling (2016) finally mapped it empirically using 80 real-world robot faces (not the usual synthetic morphs, which have their own distortions). They found the valley in explicit likability ratings and in an investment-game measure of social trust — subjects gambled less money on robots whose faces sat in the valley. So the valley is not just an intuition; it's measurable in both ratings and behavior.

The Polar Express (2004) is the classic example — pioneering motion capture that was technically impressive but emotionally disturbing. Dreamworks had to reduce the realism of Princess Fiona in Shrek because children cried during test screenings.

Now compare with a stylized animation released nearly a decade later that deliberately stayed out of the valley:

Why does the uncanny valley exist? One theory: our Bayesian visual system has learned through experience what real faces look like. When something is 99% right but 1% wrong, the mismatch triggers alarm — "something is wrong here but I can't pinpoint what." Perfect chimeric symmetry may fall in this valley because real faces are never perfectly symmetric.

Sexual dimorphism: the feature that makes supermodels super

While averageness and symmetry create a pleasant baseline, sexual dimorphism is what distinguishes a stunning face from a merely pretty one.

At puberty, testosterone stimulates jaw growth, brow ridge, and facial hair. Estrogen inhibits these features and increases lip size. These hormonal differences create the visible sexual dimorphism in adult faces.

For heterosexual males evaluating female faces: more feminine features = more attractive, with no apparent ceiling. Exaggeratedly large eyes, thin jaws, small chin (youthfulness signals) combined with high cheekbones (sexual maturity signal) = peak attractiveness.

For heterosexual females evaluating male faces: it's more complicated. Masculine features (bigger jaw, heavier brow, smaller cheeks) are preferred — but only to a point. Too-extreme masculine features are perceived as domineering and less attractive. Women seem to prefer faces with masculine features "peppered with some femininity."

The ovulatory shift hypothesis — contested

Women's preferences for masculine features may shift across the menstrual cycle. During ovulation, preferences for masculine features increase — especially for short-term sexual partners. For long-term partners, the preference stays relatively stable and favors a mix of masculine and feminine features. Penton-Voak et al. (1999) introduced this finding in Nature. The evolutionary story: reproduce with the high-testosterone male (good genes) but partner with the more nurturing, less dominant male (good father).

Whether this effect is real is one of the live arguments in the field, and it's worth slowing down on. Gildersleeve, Haselton & Fales (2014) meta-analyzed 50 studies and reported "robust cycle shifts" specifically for short-term mating cues. Wood, Kressel, Joshi & Louie (2014), running an independent meta-analysis on a partially overlapping set of 58 reports, concluded the few significant shifts were research artifacts — limited to looser definitions of the fertile phase and to published work, with the effect declining over time as methods tightened. So you have two meta-analyses, both reasonably done, reaching opposite conclusions on the same dataset. The honest read of the literature in 2024 is that if an ovulatory shift exists, it's smaller and narrower than the early studies suggested, and the burden of proof has shifted to studies with strict hormonal confirmation of the fertile window.

Face versus body: what matters when?

A study by Confer, Perilloux & Buss (2010) — mimicking the premise of the TV show Naked Attraction — asked people a blunt question: before a date, would you rather see someone's face or body first? The clever part was that participants couldn't see both. They had to choose one first, and the researchers manipulated the relationship context: short-term fling or long-term partner?

The finding: men shifted dramatically between conditions — choosing faces 75% of the time for long-term relationships but only 48% for short-term ones. Women showed much less shift (71% face for long-term, 68% for short-term).

The hypothesis: for long-term relationships, faces carry more information about personality, emotional expression, and identity. For short-term relationships, body information becomes relatively more important — especially for men.

We judge character from faces — and we shouldn't

Even though the history of physiognomy is riddled with racism and pseudoscience, modern research shows that people absolutely do make social judgments based on facial features. The judgments are consistent across people, but consistency is not accuracy.

Four properties of these judgments matter, and they are easy to confuse.

Consistency. Show many people the same face and they'll converge on similar judgments of trustworthiness, competence, and dominance.

Speed. Judgments are made in about 33 milliseconds. Todorov, Pakrashi & Oosterhof (2009) tested exposures at 17, 33, 100, and 167 ms. At 17 ms — likely below conscious face perception — participants couldn't discriminate trustworthy from untrustworthy faces. At 33 ms they could. The correlation with unlimited-time judgments climbs sharply from 33 to 100 ms, then almost levels off — meaning a third of a second of face is doing nearly all of the work of an unhurried impression.

Consequences. Todorov et al. (2005) found that the candidate who looked more "competent" won U.S. Senate and gubernatorial races — a prediction that held even when Swiss children, who had never seen the candidates, judged the faces.

Accuracy. Despite being consistent and consequential, these judgments have very small effect sizes for actual personality prediction. Olivola & Todorov (2010) analyzed over a million appearance-based judgments and found that judges relying on faces actually performed worse than they would have by ignoring appearance and using base rates. Consistency without accuracy is the danger signal here: we agree with each other on judgments that are mostly wrong, which is the worst possible combination — a coordinated bias.

The social face space

Research has identified two principal dimensions underlying face judgments:

Almost all the words people use to describe faces collapse onto these two dimensions. Attractiveness correlates strongly with the trustworthiness/valence axis. Aggressiveness is anti-correlated with trustworthiness.

Why these two dimensions? An evolutionary argument

From an evolutionary perspective, when meeting a stranger, two things matter for survival:

• Intention to cause harm → trustworthiness/valence dimension
• Ability to cause harm → power/dominance dimension

Someone who intends harm AND has the ability to carry it out is dangerous. Someone who intends no harm but is powerful is safe. This may explain why our face-reading system evolved to assess these two specific dimensions so quickly.

The overgeneralization hypothesis

If these judgments aren't accurate predictions of personality, why are they consistent? The overgeneralization hypothesis offers an answer: we're not reading personality. We're reading resemblance to emotional expressions.

"Resting bitch face" captures this. Some people's neutral facial structure resembles an angry expression — lowered brows, thinned lips. Other people's neutral face resembles a happy expression — upturned mouth corners, wide eyes. We're accurately reading the emotional resemblance, but falsely attributing it to a personality trait.

The dominance dimension might be different — it could be a more honest signal. Higher testosterone produces a larger jaw, heavier brow, more facial hair. These features are genuinely correlated with physical strength. So the "ability to cause harm" assessment may be reading a real biological signal, while the "intention to cause harm" assessment is mostly overgeneralization from emotional resemblance.

Essentialism: it's not just the face, it's the story behind it

Paul Bloom's essentialist argument adds another layer: our pleasure from things — art, faces, food — is shaped by what we believe about their origin and history.

This was demonstrated in a class experiment: students rated abstract art labeled either "from a museum" or "AI-generated." Same images. The museum-labeled ones were rated higher (the effect didn't reach statistical significance in the class sample, but it's reliable in controlled lab settings). Plassmann et al. (2008) showed the same effect at the neural level: identical wine activated more medial orbitofrontal cortex (a reward area) when subjects believed it was expensive than when they believed it was cheap. Belief restructures the experience itself.

This connects to a common experience: you meet someone you don't find attractive, but after getting to know their personality, their face literally starts looking better to you. The physical features haven't changed — your knowledge about the person has changed how your brain processes their face.

Beyond faces: what makes bodies attractive

The same two principles that predict facial attractiveness — symmetry and sexual dimorphism — also drive body preferences. But this is the section where the "universal preferences" framing fractures hardest, and the cracks are worth taking seriously.

Body symmetry is an honest signal

Fluctuating asymmetries (random deviations from bilateral symmetry in paired features like hands, feet, ears) are distinct from directional asymmetries (normal, functional differences like brain lateralization or testicle positioning). Fluctuating asymmetry honestly signals developmental instability, poor health, or inability to withstand environmental stress.

One study found that men with lower fluctuating asymmetry had more sexual partners, had sex earlier in relationships, and their female partners reported more orgasms. Women with more symmetric breasts are more fertile, and breasts become more symmetric during ovulation. Whatever else is going on, symmetry is doing real work.

The waist-to-hip ratio claim — and where it breaks

The most famous claim about body attractiveness is Singh's (1993) finding that men universally prefer a waist-to-hip ratio of ~0.7 regardless of body size. Audrey Hepburn and Marilyn Monroe — radically different body types — share that ratio. A WHR of 0.7 correlates with optimal estrogen levels, overall health, and fertility.

The "universal" part doesn't survive isolation testing. Yu & Shepard (1998) tested WHR preferences in the Matsigenka of southeastern Peru — a population substantially more isolated than the Aché. Matsigenka men preferred higher WHRs (closer to 0.9 — an almost tubular shape), describing it as healthier. In tests of Matsigenka villagers along a gradient of increasing Western contact, the preferred WHR shifted toward the Western 0.7 with exposure. Wetsman & Marlowe (1999) replicated the direction with the Hadza of Tanzania, who also preferred WHRs closer to 0.9. The original "universal 0.7" was — at least in part — a result of testing populations already in the orbit of Western media.

This doesn't mean WHR is meaningless. It means the specific preferred ratio is environment-tuned, while the use of waist-to-hip information as a signal of health and reproductive value may be more universal. Biology gives you the question; environment fills in the answer.

Fat countries prefer thin; thin countries prefer fat

The cleanest cultural effect on body preference: Swami et al.'s International Body Project surveyed over 7,400 people across 10 world regions and found that countries with higher BMI prefer thinner women, and countries with lower BMI prefer heavier women. The environmental security hypothesis explains this — in environments where survival is uncertain, weight signals maturity and proven survival. An older, heavier person has a track record of not dying.

This shows up in surprising data: the Playboy Playmate of the Year during economic downturns tends to be older, heavier, taller, with a larger waist and higher BMI — all signals of maturity and security. Movie actresses follow the same pattern: more youthful "baby features" in good economic times, more mature features during recessions.

The halo effect: beauty's unfair advantage

The halo effect — where one positive trait colors all other judgments — means that physical attractiveness has measurable real-world consequences:

• Hamermesh & Biddle (1994) found that "plain" people earn 5–10% less than people of average looks, who earn less than the good-looking — a labor-market penalty roughly the size of a year of college.
• Stewart (1980) observed 67 actual criminal defendants and found attractiveness predicted both minimum and maximum sentences at p-values below .001 — more attractive defendants drew shorter sentences for comparable crimes.
• Eberhardt et al. (2006) "Looking Deathworthy" is the most uncomfortable finding. In 44 Black-defendant capital cases in Pennsylvania involving white victims, the more stereotypically Black a defendant's features, the more likely the death penalty. Face-based judgments don't just nudge outcomes; in capital cases they help decide who dies.
• Teachers rate attractive children as better students (which may become self-fulfilling).
• Political candidates judged as more "competent-looking" win elections more often.

The original Nisbett & Wilson (1977) study showed the effect works in reverse too: a professor who acted warm was rated as physically more attractive than the same professor acting cold. Subjects didn't realize their personality impression was coloring their attractiveness judgment.

Assortative mating: we match on attractiveness

People don't pair randomly. The correlation between partners' attractiveness levels is 0.39–0.49 — almost as strong as the correlation between parent height and child height. We assortatively mate on attractiveness, and also on political values, religiosity, intelligence, and even neuroticism.

The "hot or not" research clarified the mechanism. Hitsch, Hortaçsu & Ariely (2010) analyzed an online dating dataset and showed it's market dynamics, not genuine preference for similar attractiveness. People at all attractiveness levels prefer the most attractive partners — but their dating behavior (who they message, who they pursue) is constrained by their own attractiveness level. The preference curve keeps going up; only the action curve peaks near your own level.

A counterargument worth taking seriously: maybe the "biology" isn't adaptive

Everything I've written so far leans on a particular framing — that the underlying biological signal is doing fitness-relevant work: signaling health, fertility, developmental stability, reproductive age. This is the honest signaling view, and it's the dominant one in evolutionary psychology.

It is not the only serious view. Richard Prum's The Evolution of Beauty argues that much of sexual selection produces arbitrary aesthetic preferences that have no honest-signal function at all. They evolve through what Prum calls the Lande–Kirkpatrick null model: a positive feedback loop where a preference and the trait it favors co-evolve simply because females who prefer the trait have sons with the trait who get preferred. No fitness benefit required. Beauty Happens, in Prum's phrase.

If Prum is right, even the "biological signal" half of the biology-vs-culture story I've been telling is partly arbitrary — a frozen accident of mate choice, not a window into reproductive value. The honest-signaling framework would still explain some preferences (extreme asymmetry really does track developmental damage), but the universal preference for large eyes and small chins would not need to mean those features signal fitness. They could just be what got locked in.

I don't think Prum's framework dissolves the case I've made. The cross-cultural agreement, the infant preferences, the consistency of the youthfulness markers across populations — these are still real and still better explained by a shared underlying preference than by independently invented cultural conventions. But the interpretation of what that shared preference means is more open than the standard evolutionary-psychology story suggests. "Built in" doesn't necessarily mean "tracking reproductive value." It might just mean "locked in long ago for reasons that no longer apply."

TLDR

• Babies as young as 3 months look longer at faces adults rate as attractive. They've had time to learn the mother's face, but not the structural preference for averaged, prototypical faces of strangers, different races, and different sexes.
• Cross-cultural agreement on attractive faces is robust (r ≈ .93 across White, Asian, and Hispanic raters; near-universal preference for large eyes and delicate jaws). The "isolated tribes" framing oversells the isolation, but the agreement on youthfulness markers survives the critique.
• Culture amplifies biology into supernormal stimuli — Tinbergen's herring-gull chicks pecking at giant fake red dots scaled up to BBL surgery, Disney eyes, and bodybuilding extremes. The biological signal was honest. The cultural amplification rarely is.
• Galton's 1878 attempt to identify a "criminal face" accidentally discovered the most robust finding in the field: averaged faces are attractive. Supermodels deviate from the average in specific ways — youthfulness markers plus puberty markers, a biologically impossible combination.
• The uncanny valley is now empirically confirmed (Mathur & Reichling 2016), not just Mori's intuition — perfect symmetry and near-but-not-quite-human faces produce measurable revulsion in both ratings and trust-game behavior.
• The famous ovulatory-shift hypothesis is contested — two meta-analyses (Gildersleeve 2014, Wood 2014) reached opposite conclusions on overlapping data. If the effect is real, it's smaller than early studies suggested.
• Face-based personality judgments are made in 33 ms, are extremely consistent across people, predict election outcomes and criminal sentences — and are less accurate than ignoring the face and using base rates. Consistency without accuracy is the worst combination: coordinated bias.
• The "universal" WHR of 0.7 doesn't survive isolation testing — Matsigenka and Hadza men prefer WHRs closer to 0.9. The specific preferred ratio is environment-tuned; the use of waist-to-hip information as a signal may be more universal.
• The halo effect costs "plain" people 5–10% in wages, predicts criminal sentences, and in capital cases helps decide who is sentenced to death (Eberhardt 2006). Beauty's unfair advantage is not a metaphor; it's measurable.
• Even the "biology" half of the story is contested. Richard Prum's arbitrary-aesthetic-evolution framework argues that much of what looks like an honest fitness signal is actually a frozen accident of mate choice. The data this post leans on doesn't distinguish the two interpretations as cleanly as the honest-signaling story claims.

The line that stays with me. Beauty feels like the most personal taste you have. It's mostly the least personal — newborns and Paraguayan foragers share most of it with you. But "shared" doesn't have to mean "adaptive." The features we call beautiful may be tracking real biological information, or they may be tracking nothing at all except a long evolutionary feedback loop that locked in early. Either way, the most honest thing to ask when you find yourself strongly drawn to a face isn't "are they attractive?" It's "what features am I overgeneralizing from, and what would I think if I'd grown up somewhere the answers were different?"

Sources:

• Bar, M., Neta, M., & Linz, H. (2006). Very first impressions. Emotion, 6(2), 269–278. — 39 ms threat-judgment threshold.
• Barrett, D. (2010). Supernormal Stimuli: How Primal Urges Overran Their Evolutionary Purpose. W. W. Norton. — Extension of Tinbergen's concept to humans.
• Bushnell, I. W. R., Sai, F., & Mullin, J. T. (1989). Neonatal recognition of the mother's face. British Journal of Developmental Psychology, 7, 3–15. — Newborns recognize the mother's face within hours.
• Chatterjee, A. (2014). The Aesthetic Brain: How We Evolved to Desire Beauty and Enjoy Art. Oxford University Press.
• Confer, J. C., Perilloux, C., & Buss, D. M. (2010). More than just a pretty face: men's priority shifts toward bodily attractiveness in short-term versus long-term mating contexts. Evolution and Human Behavior, 31(5), 348–353.
• Cunningham, M. R., Roberts, A. R., Barbee, A. P., Druen, P. B., & Wu, C. H. (1995). "Their ideas of beauty are, on the whole, the same as ours." Journal of Personality and Social Psychology, 68(2), 261–279.
• Eberhardt, J. L., Davies, P. G., Purdie-Vaughns, V. J., & Johnson, S. L. (2006). Looking deathworthy. Psychological Science, 17(5), 383–386.
• Etcoff, N. (1999). Survival of the Prettiest: The Science of Beauty. Anchor Books.
• Galton, F. (1878). Composite portraits. Nature, 18, 97–100. — The accidental discovery that averages are attractive.
• Gildersleeve, K., Haselton, M. G., & Fales, M. R. (2014). Do women's mate preferences change across the ovulatory cycle? A meta-analytic review. Psychological Bulletin, 140(5), 1205–1259.
• Hamermesh, D. S., & Biddle, J. E. (1994). Beauty and the labor market. American Economic Review, 84(5), 1174–1194.
• Hill, K., & Hurtado, A. M. (1996). Aché Life History: The Ecology and Demography of a Foraging People. Aldine de Gruyter. — Primary ethnographic source on Aché contact history.
• Hitsch, G. J., Hortaçsu, A., & Ariely, D. (2010). What makes you click? — Mate preferences in online dating. Quantitative Marketing and Economics, 8(4), 393–427.
• Jones, D., & Hill, K. (1993). Criteria of facial attractiveness in five populations. Human Nature, 4(3), 271–296.
• Langlois, J. H., Ritter, J. M., Roggman, L. A., & Vaughn, L. S. (1991). Facial diversity and infant preferences for attractive faces. Developmental Psychology, 27(1), 79–84.
• Langlois, J. H., & Roggman, L. A. (1990). Attractive faces are only average. Psychological Science, 1(2), 115–121.
• Langlois, J. H., et al. (2000). Maxims or myths of beauty? Psychological Bulletin, 126(3), 390–423.
• Mathur, M. B., & Reichling, D. B. (2016). Navigating a social world with robot partners: a quantitative cartography of the uncanny valley. Cognition, 146, 22–32.
• Mori, M. (1970/2012). The uncanny valley. Energy, 7(4), 33–35. English translation: IEEE Spectrum, 2012.
• Nisbett, R. E., & Wilson, T. D. (1977). The halo effect: evidence for unconscious alteration of judgments. Journal of Personality and Social Psychology, 35(4), 250–256.
• Olivola, C. Y., & Todorov, A. (2010). Fooled by first impressions? Reexamining the diagnostic value of appearance-based inferences. Journal of Experimental Social Psychology, 46(2), 315–324.
• Pascalis, O., de Haan, M., & Nelson, C. A. (2002). Is face processing species-specific during the first year of life? Science, 296(5571), 1321–1323.
• Penton-Voak, I. S., et al. (1999). Menstrual cycle alters face preference. Nature, 399, 741–742.
• Plassmann, H., O'Doherty, J., Shiv, B., & Rangel, A. (2008). Marketing actions can modulate neural representations of experienced pleasantness. PNAS, 105(3), 1050–1054.
• Prum, R. O. (2010). The Lande–Kirkpatrick mechanism is the null model of evolution by intersexual selection. Evolution, 64(11), 3085–3100.
• Prum, R. O. (2017). The Evolution of Beauty: How Darwin's Forgotten Theory of Mate Choice Shapes the Animal World — and Us. Doubleday. — The arbitrary-aesthetic-evolution counterargument.
• Rhodes, G. (2006). The evolutionary psychology of facial beauty. Annual Review of Psychology, 57, 199–226.
• Singh, D. (1993). Adaptive significance of female physical attractiveness: role of waist-to-hip ratio. Journal of Personality and Social Psychology, 65(2), 293–307.
• Stewart, J. E. (1980). Defendant's attractiveness as a factor in the outcome of criminal trials. Journal of Applied Social Psychology, 10(4), 348–361.
• Swami, V., et al. (2010). The attractive female body weight and female body dissatisfaction in 26 countries across 10 world regions. Personality and Social Psychology Bulletin, 36(3), 309–325.
• Tinbergen, N. (1948). Social releasers and the experimental method required for their study. Wilson Bulletin, 60(1), 6–51.
• Todorov, A., Mandisodza, A. N., Goren, A., & Hall, C. C. (2005). Inferences of competence from faces predict election outcomes. Science, 308(5728), 1623–1626.
• Todorov, A., Pakrashi, M., & Oosterhof, N. N. (2009). Evaluating faces on trustworthiness after minimal time exposure. Social Cognition, 27(6), 813–833. — The 33 ms judgment threshold.
• Tovée, M. J., et al. (2006). Changing perceptions of attractiveness as observers are exposed to a different culture. Evolution and Human Behavior, 27(6), 443–456.
• Wetsman, A., & Marlowe, F. (1999). How universal are preferences for female waist-to-hip ratios? Evidence from the Hadza of Tanzania. Evolution and Human Behavior, 20(4), 219–228.
• Willis, J., & Todorov, A. (2006). First impressions: making up your mind after a 100-ms exposure to a face. Psychological Science, 17(7), 592–598.
• Wood, W., Kressel, L., Joshi, P. D., & Louie, B. (2014). Meta-analysis of menstrual cycle effects on women's mate preferences. Emotion Review, 6(3), 229–249.
• Yu, D. W., & Shepard, G. H. (1998). Is beauty in the eye of the beholder? Nature, 396, 321–322.