Abstract: Because different regions of the synthetic aperture radar (SAR) image have different statistical characteristics, the phenomenon of missing details and “ghost” artifacts often occur in the traditional nonlocal SAR image despeckling method. As a result, a classificationbased SAR image despeckling method is proposed. Firstly, the image is classified into homogeneous regions and heterogeneous regions through comparing the variation of average coefficient with the threshold value. Secondly, according to the classification of the central pixel of the target block, the appropriate parameters are selected to perform block matching. Finally, the target block located in homogeneous regions is proceed by weighted average with similar blocks of the target block. On the other hand, the target block located in heterogeneous regions is performed by 3D hardthresholding shrinkage with similar blocks of the target block. The proposed method can not only improve the accuracy of block matching through adjusting the block size and the search window size on the basis of the characteristics of image blocks, but also achieve a satisfied balance between smoothing (in homogeneous region) and feature preservation (in heterogeneous region) by selecting different methods to deal with different regions. The experiment results show that the proposed method makes full use of the characteristics of homogeneous and heterogeneous regions, which combine the advantage of different methods to improve the despeckling performance and reach a high level in terms of both performance index and visual quality.