Fold a photograph of a face down the middle and mirror one half onto the other, and the result looks slightly haunted rather than improved. Everyone has tried this trick at some point, usually on their own face. The reaction is never delight.

Symmetry has been the standard answer to what makes a face attractive for about three decades, repeated in magazine listicles and first-year psychology lectures with equal confidence. A large cross-cultural study has now run that answer through a much harder test than it normally gets, and symmetry did not come out the other side.

What the researchers measured

Karel Kleisner of Charles University in Prague led a multi-country team that gathered 1,550 photographs of faces from ten populations: Brazil, Cameroon, Czechia, Colombia, India, Namibia, Romania, Turkey, the UK and Vietnam. Every image was unretouched, shot front on with a level head and a direct gaze, and hand marked with 72 anatomical landmark points.

Three separate things were then pulled out of those landmarks. Symmetry. Sex typicality, meaning how masculine or feminine the underlying shape reads. And distinctiveness, which is just how far a face sits from the mathematical average of its own population.

Attractiveness ratings came from opposite-sex volunteers in the subject’s own country, which matters more than it sounds. A Namibian face was judged against Namibian expectations rather than a global template built mostly out of European faces.

Symmetry did not survive the analysis

In a paper published in Evolution and Human Behavior, the team reported that distinctiveness pulled attractiveness ratings down in every one of the ten populations, with no exceptions and no cultural wobble. Femininity lifted ratings of women’s faces. Masculinity did close to nothing for men’s.

Symmetry, once the other factors were in the model, registered as noise.

In coverage by PsyPost when the findings landed, that result cut against a claim most people treat as settled. And the framing in the paper is careful: what the raters preferred was proximity to a local mean, not balance.

Why earlier studies said the opposite

Older experiments tended to manufacture their own faces. A researcher takes one photograph, digitally warps it into a more symmetrical version and a less symmetrical version, then asks people to choose. Symmetry wins comfortably under those conditions, which is roughly how David Perrett and colleagues at St Andrews reached their much-cited 1999 conclusion that symmetrical faces are preferred.

Trouble is, a warped pair of images differs on exactly one dimension. Real faces differ on dozens simultaneously, and a person glancing at one has far more information competing for attention. Manipulated stimuli reveal the cues a viewer is capable of picking up, which is a different question from what people lean on in the wild.

A second team ran it again

A team led by Benedict Jones at the University of Strathclyde in Glasgow published a follow-up in Scientific Reports in February 2025. Fifty male photographs and fifty female ones, none of them digitally altered, went out to 400 volunteers, with shape measured off the images instead of engineered into them.

Averageness predicted the ratings. So did femininity, in women’s faces only. Masculinity and symmetry predicted nothing at all.

Two independent samples landing on an identical pattern are considerably more interesting than one doing so.

Both groups are also upfront that the effects are weak. Averageness accounts for a sliver of why one face rates higher than another, and the remainder is skin, colouring, expression, body weight and a pile of things no one has finished measuring.

What average is actually doing

Averageness is a peculiar word to use as a compliment. The idea traces back to Judith Langlois and Lori Roggman, who in 1990 digitised student photographs, blended them by pixel, and found the composites outscored almost every individual face fed into them. The more faces in the blend, the higher the rating climbed.

Two explanations have been competing ever since. One holds that a prototypical face advertises a solid immune system and decent genes. The other holds that brains process familiar shapes quickly and then mistake that ease for pleasure. Neither has won, and both papers here name the question as unresolved.

None of this argues that asymmetry goes unnoticed. A drooping eyelid or a jaw knocked off line in a bike accident is obvious to anyone looking. The finding is narrower and stranger than that: within the ordinary range of wonk that almost every human face carries, the wonk does not appear to shift the score.

What has shifted is the reference point. The average doing the work is local, drawn from the faces someone grew up surrounded by. Beauty standards travel a good deal less than the marketing implies.

Which puts a fresh and slightly awkward light on everybody who has been told, warmly, that they have a really interesting face.