Previous work has shown that principal component analysis (PCA) of three-dimensional face models can be used to perform recognition to a high degree of accuracy. However, experimentation with two-dimensional face images has shown that PCA-based systems are improved by incorporating linear discriminant analysis (LDA), as with Belhumier et al’s fisherface approach. In this paper we introduce the fishersurface method of face recognition: an adaptation of the two-dimensional fisherface approach to three-dimensional facial surface data. Testing a variety of pre-processing techniques, we identify the most effective facial surface representation and distance metric for use in such application areas as security, surveillance and data compression. Results are presented in the form of false acceptance and false rejection rates, taking the equal error rate as a single comparative value.