To solve the problem of automatic target segmentation for SAR images, a target segmentation algorithm for SAR images was proposed after the analysis of nonsubsampled contourlet transform and pulse coupled neural networks. Via researching the characteristics of low and high frequency, the conclusion was acquired that the first one contained probable region of target mainly. Correspondingly, the latter contained fine contour and background disturbance mainly. Fire image of low frequency was produced by Pulse Coupled Neural Networks (PCNN) acting on low frequency image, and the region which the target located was confirmed on the basis of segmentation for the fire image using OTSU method. A characteristic figure was constructed for the high frequency, and the fine configuration of the target was acquired on the basis of segmentation for characteristic figure’s fire image. Experiments with MSTAR images were processed and the proposed algorithm was compared with algorithms based on fuzzy C mean and Markov random fields. The results indicate that the proposed algorithm which has more accurate segmentation for SAR target and more strongly immune ability for speckle was effective.