Face recognition is used to identify a person in a given image using existing images in the database. Principal component analysis reduces the high dimensional data (2-D image represented as 1-D vector) to a low dimensional feature space by using the eigen value decomposition of covariance matrix of the training data. In this project, Yale Face Database was used for the experiments. Using eigenvectors corresponding to k highest eigenvalues (k«dimensionality of 1D image vector), each image was projected to k-dimensional space (eigenface). The test image was also projected to the k-dimensional space using eigenvectors obtained from training set and was classified as belonging to the training sample which is closest to it based on Euclidean distance. Test data comprised of images not in the training set including images of person in the training set, person not in the training set and some non-face image. In addition, image compression using Singular Value Decomposition was studied and the performance was analyzed in terms of compression factor, mean square error and peak signal to noise ratio.