For the issue that performance of face recognition algorithms was impacted by occlusion, a robust face recognition algorithm by using fast-weighted principal component analysis (FW-PCA) detecting occluded region was proposed. FW-PCA was used to detect occluded region, and occluded region of input images were compared with gallery images. Local binary pattern (LBP) was used to determine the optimal weights and phase-only correlation (POC) was used to get occluded mask. Matching score of each image was calculated, face recognition was finished by nearest neighbor classifier. Experimental results on FRGC2 and UND show that the recognition accuracy can achieve 99.6%. It has better recognition performance than several advanced recognition algorithms.