Abstract
In one of the most replicated findings in aesthetic perception, faces that are mathematically averaged from many individuals are rated as more attractive than the constituent faces. The effect is robust across cultures, age groups, and methodologies — but it’s smaller than the headline suggests, and it explains less variance in attractiveness than people assume.
How averageness is measured
A composite face is built by aligning landmarks across a set of faces and averaging pixel positions and intensities. The resulting “mean shape” face is more attractive than the median individual face but, importantly, not the most attractive face — high-symmetry, high-dimorphism individuals can rate above the composite.
Why averageness is attractive
The dominant explanation is prototype proximity: averaged faces sit closest to the cognitive prototype an observer has built up from many seen faces. They’re also necessarily smoother (averaging cancels out skin imperfections) and more symmetric (averaging cancels out individual asymmetries).
What averageness doesn’t capture
Averageness is necessarily aggregating, so it predicts what observers prefer on average, not what any specific observer prefers. Observers with high specificity in their preferences (a “type”) can find non-average faces preferable. Cross-cultural cohorts complicate the picture: average for whom?
Practical takeaways
VERASOMA compares your geometry against demographic cohorts (age band × sex × ethnicity) rather than a global mean. A score of 70+ on cohort averageness is typical and meaningful; the headline effect of “more average is more attractive” is real but should be interpreted in that cohort-specific context.
FAQ
If average is most attractive, why are unusual faces sometimes preferred? Averageness explains a meaningful but bounded share of variance. Individual preferences can lean toward non-average features (a strong jaw, large eyes), which average composites can’t reproduce.
Does averageness equal “boring”? No — averaged faces lose imperfections, not character.