A Random Sampling Subspace Locality Preserving Projection(RSSLPP) method is proposed to improve the recognition performance of a single Locality Preserving Projection(LPP).At the training stage,based on random sampling of the principle component subspace of training set,the multiple discrepant and complementary LPP subspaces are generated by LPP method.At the recognition stage,test sample is successively projected into each random sampling principle component subspace,then the nearest neighbor classifier is used for classification and recognition.Finally,majority voting criterion is used to fuse the recognition result of each LPP subspace.The experimental results on FERET subset illustrate that the performance of RSSLPP method is superior to those of Eigenface,Fisherface,LPP and discriminant LPP(DLPP).The recognition accuracy of RSSLPP is over 10% higher than that of LPP.The RSSLPP method can effectively combine the complementary information of each LPP subspace and can improve face recognition accuracy.