Abstract
People from different cultural and economic backgrounds agree about which faces are attractive, suggesting that humans share an attractiveness “template.” Much research in the last two decades has focused on the composition of this template, such as whether it is primarily sensitive to facial averageness, symmetry, or sexual dimorphism. However, we have little understanding of how different aspects of appearance codetermine perceptions of attractiveness. Using a simple neural network model, we can reconstruct an attractiveness template from images of faces and attractiveness judgements of those images. The model solves the difficult problem of replicating and predicting human attractiveness judgements to images of faces. It reduces judgements into simpler image-based factors that are sensitive to aspects of facial appearance. These factors are strikingly similar to averageness, symmetry, and sexual dimorphism.