Abstract
Previous research on facial attractiveness has shown that mathematically average faces are perceived as highly attractive. In this study, we obtained attractiveness ratings for 2000 male and 2000 female faces sampled from a 50 dimensional face space. This face space approximates the shape and reflectance variance in human faces. After collecting the ratings, we used second-order polynomial regression to create an attractiveness function. This data-driven approach allows us to predict the attractiveness of any arbitrary face. The attractiveness function shows that while averageness is important for some dimensions, it is not for others. In particular, attractive male faces have darker skin, darker eyebrows, more beard, and longer jaws than the average male face. Attractive females have upper and lower eyelids that are much darker than those of the average female face. For many other dimensions, however, the theoretically most attractive female is near the mean. Additionally, the attractiveness function confirms the importance of sexual dimorphism for some, but not all dimensions.