In this paper, a novel SVD-based method is developed to classify face images. The experimental results show that the novel method outperforms the existing SVD-based face recognition methods. In addition, we investigate the performances on face recognition of singular values, left and right singular vectors generated from SVD and assess the existing SVD-based face recognition methods. It appears that by compared with singular values and singular vectors, SVD-based reconstruction images are more useful in classifying face images. In practice, through the SVD-based image reconstruction process, we may weaken the side effects on face classification of varying imaging conditions and facial expressions.