The problems of strong coherent speckle noise and the lack of prior knowledge of background and targets, make segmentation difficult in synthetic aperture radar(SAR) images. A segmentation algorithm based on improved fuzzy C-means (FCM) clustering and Markov random field (MRF) is proposed. Firstly, the efficiency of fast FCM is improved by the adaptive nonlocal mean filter and the initial clustering center selection rule based on histogram peak points. Secondly, the SAR image is segmented by improved FCM and MRF respectively, and the optimal segmentation regions are adaptively selected by constructing the joint membership matrix. Finally, the final segmentation result is optimized by morphological operations. The experimental results show that the proposed algorithm has better antinoise performance and can segment multiclass SAR images quickly and efficiently.