As for Synthetic Aperture Radar (SAR) target recognition,a method based on modified sparse representation is proposed.Firstly,based on traditional Sparse Representation-based Classification (SRC),the sparse representation coefficient vector is calculated on the global dictionary.Based on this,the optimal local dictionaries are selected for different training categories.Then,the test samples are reconstructed using different local dictionaries.Finally,the target categories of the test samples are determined by comparing the reconstruction errors of different categories.In the experiments,the MSTAR dataset is taken as samples for testing and validation.The superiority of the performance of the proposed method is confirmed.