It is important to extract discriminant features for pattern recogniti on.Manifold learning can deal with the nonlinearity hidden in the data and the sparse representation shows its robustness. To extract discriminant and robust features,in this paper,a method called loca l sparse representation and discriminant analysis is proposed,which preserves the local sparse relationship and maximizes the inter-class separatability.As a result,the extra cted features are sparse,discriminative and helpful for classification.Experiments on two ope n face databases, ORL and Yale,show that the proposed algorithm improves the accuracy,and the corr ectness and effectiveness of the proposed algorithm are confirmed.