Abstract: In order, to reduce the impact environmental requirements and overcome the effects of illumination on face recognition, a novel recognition method is presented based on normalization illumination in frequency-domain, via analyzing the Amplitude-frequency and phase-frequency characteristics of human faces. Collected images under any light conditions are normalized so that the light is exactly the same for the training images, while retaining the distinctive features of human faces. As the different information of faces is usually small, the smallest non-zero eigenvector is chosen as a facial feature. Experimental results show that our method is more robust against light than traditional methods.