This paper proposes a nonlocal means filter-based speckle reduction algorithm for synthetic aperture radar (SAR) images.In contrast to traditional methods，the proposed method is capable of discerning pixels of same structural features by exploiting similarity between image patches，leading to efficient reduction of structural information in despeckling.In terms of characteristics of similarity of various structural primitives，a two-stage nonlocal means filtering is used for alleviating the impact of noise on structural similarity measures，and hence improve the despeckling performance.The proposed method has been evaluated on simulated SAR images with varying levels of speckle noise and a real SAR image.Compared to the traditional Lee filter，enhanced Lee filter and the speckle reducing anisotropic diffusion (SRAD)，the proposed method substantially improves the despeckling performance especially on the preservation of structure information.