Abstract: We present a multi-pose face recognition method based on two-factor analysis model.The method refines the traditional two-factor analysis model,and partly solves the problem that facial feature is sensitive to the pose variation.Large number of 3D face data is trained in the two-factor analysis model to get robust and differential pose factors.In the experiment of FERET facial database,the best recognition accuracy is 92.5%.The results show that both the global and local facial features maintain a good pose variation robustness and high significance of feature descriptor after the pose factor separation.