How is attractiveness measured?

The basic design has barely changed in decades. A set of face photos, taken in the same light and pose, is shown to a group of raters who score each one. Individual raters disagree, but the average of many raters is stable, and different groups tend to rank the same faces in a similar order. That average is the working measure of attractiveness.

The next step is to look for what predicts it. Researchers alter faces, for example making them more symmetrical, blending them towards the population average or evening out skin tone, and see whether ratings move. Reviews of this work find several consistent signals: facial symmetry, averageness, even skin colour and texture, and a lean but healthy amount of facial fat.

The method has limits worth knowing. It measures still photos, mostly of young adults, judged by strangers at first sight, so it says little about how someone reads in motion, with expression, or to people who know them. Shared signals explain part of the agreement between raters; individual taste explains the rest. And no study has shown that a single ratio, golden or otherwise, is a law of beauty. Some of the signals are workable, too: skin evenness and facial fat respond to sleep, sun protection and body composition, while symmetry and bone structure largely do not.

Measurement tools such as the facial harmony test measure geometry: distances and proportions between landmarks. That is a real measurement, but of proportion, not of beauty. AI raters give a structured opinion instead, and the useful ones say where their average sits and grade the photo first. Is an AI face rating accurate covers how to judge one, and is facial harmony objective covers the proportion side. The attractiveness test explains which of these signals one photo can support.