In this paper, we propose a new method for face recognition under varying illumination conditions using a single input image. Our method is based on a statistical shape-from-shading method which combines the strengths of the Lambertian model and statistical information obtained from a large number of images of different people under varying illumination. The main advantage of our method over the previous methods is that our method explicitly incorporates a correlation between surface points on a face in the MAP estimation of surface normals and albedos, so that a new image of the same face under novel illumination can be synthesized correctly even when the face is partially shadowed. Furthermore, our method introduces pixel grouping and reliability measure in the MAP estimation in order to reduce computational cost while maintaining accuracy. We demonstrate the effectiveness of our proposed method via experiments with real images.