Abstract: Human face recognition is an important technique that can be widely employed in many practical applications, and, specifically, recognizing human face under non-uniform illumination condition with unknown-position light source is a particul arly difficult problem. In this paper, a human-face-model based approach to rest ore the illumination of human face under the above condition is proposed, which work by recovering the intensity of the shaded areas in human face. Human face t exture is represented as a linear combination of several training images, and, b ased on the face texture in the bright area, face illumination is restored by re covering face texture through a least-square estimation. Experiments show that t he method is not only efficient but also fast in restoring the shaded area on hu man face.