In this paper, we study face recognition using Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) under illumination variations. A Modified Census Transform (MCT) is applied as preprocessing step to compensate illumination variations, and then PCA and LDA are employed to find lower-dimensional subspaces for face recognition. Distances between training and testing images are measured by three metrics (L1, L2, and cosine). The aim of this paper is to compare the results of two most popular subspace projection methods under illumination variation conditions.