A novel Patched Locality Preserving Projections for 3D face recognition was presented in this paper. In this paper, we firstly patched each image to get the spatial information, and then Gabor filter was used extract intrinsic discriminative information embedded in each patch. Finally Locality Preserving Projections, which was improved by Principle Components Analysis, was utilized to the corresponding patches to obtain locality preserving information. The feature was constructed by connecting all these projections. Recognition was achieved by using a Nearest Neighbor classifier finally. The novelty of this paper came from: (1) The method was robust to changes in facial expressions and poses, because Gabor filters promoted their useful properties, such as invariance to rotation, scale and translations, in feature extraction; (2) The method not only preserved spatial information, but also preserved locality information of the corresponding patches. Experiments demonstrated the efficiency and effectiveness of the new method. The experimental results showed that the new algorithm outperformed the other popular approaches reported in the literature and achieved a much higher accurate recognition rate.