Abstract: In order to overcome the limitation that two-dimensional locality preserving projection (2DLPP) needs more data to represent face features, a new method, named two-dimensional locality preserving projection-principal component analysis (2DLPP-PCA), was proposed. By simultaneously considering 2DLPP and 2DPCA, the 2DLPP-PCA can not only reduce the data needed in preserving face features, but also effectively extract the local structure information from 2DLPP and the global structure information from 2DPCA. The experiments on the ORL, Yale and CAS-PEAL-R1 face databases indicate that the 2DLPP-PCA is a high-performance method for face feature extraction, with the best average recognition rate higher than 99% when the number of training samples on the ORL face database is 6.