The general face recognition methods mostly use positive face images or with a small angle of deflection. Such databases are well established. But in real life, face recognition problems often do not happen to be positive face state, especially in monitoring and public security monitoring. The existing effect of multi-angle face recognition is not ideal, and this paper proposes a face recognition method based on Gaussian Markov Random Fields (GMRF). GMRF model is a statistical probability model that can effectively extract image texture information, which first divides the face image into several sub-blocks in different chunks, and then, for each sub-block under the block mode, extracts the GMRF feature after wavelet transformation; finally combines the GMRF features of different blocking methods, and then classifies the SVM of the Gaussian nuclear function. Experiments were carried out on the self-built data set, and the proposed method reached 98.83% for face recognition.