In this paper, a discriminative manifold learning method for face recognition is proposed which achieved the discriminative embedding the high dimensional face data into a low dimensional hidden manifold. Unlike the recently proposed LLE, Isomap and Eigenmap algorithms, which are based on reconstruction purpose, our method uses the RCA algorithm to achieve nonlinear embedding and data discrimination at the same time. Also, the LLE and Isomap algorithms are crucially depends on the appropriateness of the neighborhood construction rule, in this paper, a CK-nearest neighborhood rule is proposed to achieve better neighborhood construction. Experimental results indicate the promising performance of the proposed method.