Abstract: We present a fast and effective segmentation algorithm for MSTAR SAR target chips.The algorithm is based on finite mixture models,by which the segmentation is posed as an inference problem through introducing an improved potts model base on MRF.The final segmentation result is obtained by fast and robust parameters estimation based on combining expectation maximization and graph cut optimization.We compare its performance to conventional MRF methods based on three standard image segmentation indices using MSTAR SAR data sets,and experimental results show that our proposed algorithm has better performance than traditional MRF method.