Abstract: In this study, a novel subspace learning method named Neighborhood Preserving Fisher Discriminant Analysis (NPFDA) is proposed for face recognition. Based on Fisher Discriminant Analysis (FDA), NPFDA takes into account the local geometry structure information, changes the objective function. Thus, two abilities of manifold learning and classification are combined into the proposed method. In order to improve the discriminating power, Schur-decomposition is used to get the orthogonal basis vectors. Experimental results on the Yale face database and Feret face database demonstrate the effectiveness of the proposed method.