Gabor wavelet is a multi-resolution description. It could extract the gray feature of facial area using amplitude coefficients of Gabor wavelet even when illumination changes. Based on the fact, this paper proposes a face detection method employing BP network combined with Gabor wavelet transform. Sample features are represented by Gabor description for BP network training. First we use standard modules to compute the similarity between the modules and the picture, and get the possible facial areas. On the other hand, we use BP network to classify the possible areas and label the final result. Compared with conventional method using gray scale to train neural network, the method proposed in this paper produces better results.