Recently, some research efforts have shown that face images possibly reside on a nonlinear sub-manifold. Though Laplacianfaces method considered the manifold structures of the face images, it has limits to solve face recognition problem. This paper proposes a new feature extraction method, Two Dimensional Laplacian EigenMap (2DLEM), which especially considers the manifold structures of the face images, and extracts the proper features from face image matrix directly by using a linear transformation. As opposed to Laplacianfaces, 2DLEM extracts features directly from 2D images without a vectorization preprocessing. To test 2DLEM and evaluate its performance, a series of experiments are performed on the ORL database and the Yale database. Moreover, several experiments are performed to compare the performance of three 2D methods. The experiments show that 2DLEM achieves the best performance.