A new method, which combines multiscale technoogy, mixture model and self-organizing mixture network, is proposed for the segmentatioin of SAR image. Firstly,multiscale characteristic is distilled based on the multiscale model of SAR image. Then, the mixture model is found for the multiscale information, and its parameters are estimated by self-organizing mixture network. Lastly,Bayesian classifier is used to segment the SAR image. The experimental results shows the proposed method sufficiently captures the statistical information in a multiscale sequence of SAR image, which is then used to implement the segmentation of SAR image via multiresolution mixture algorithm and self-organizing mixture network, and the perforrnance of segmentation is obviously improved.