This paper proposes a synthetic aperture radar (SAR) target recognition method based on combination of global and local representations. Sparse representation over the global dictionary could effectively compares the relative description capabilities of different classes for the test sample. However, local dictionarybased sparse representation reflects the absolute description ability of each category on the test sample. Therefore, the two representations could complement each other to provide more information for correct decisions. The decision value vectors (i.e., reconstruction errors) from the global and local representations are fused by DempsterShafer (DS) evidence theory for robust target recognition. Experiments are conducted on public moving and stationary target acquisition and recognition (MSTAR) dataset to be compared with other SAR target recognition methods. The experimental results show the effectiveness of the proposed method.