The paper provides novel approaches for the employment of spectral information when pursuing face recognition. We designed and tested Eigenface based algorithms that improve face recognition through feature extraction, i.e. extracting the 'best bands' according to various criteria such as decorelation and statistical independence. Eigenfaces correspond to principal components and have previously been used for regular grayscale and color images. In this paper we expand theiruse to hyperspectral imagery, i.e. data sets of images of the same scene associated to narrow wavelength intervals. Our approach is a two decomposition process. In the first, the hyperspectral data is reduced to grayscale using Principal Component Analysis. In the second, the grayscale images are processed using the classical Eigenface detection algorithm. The results suggest that spectral imaging improves face classification over its counterpart.