Abstract: This paper is based on the gray-level co-occurrence matrix method, and particularly study some texture features used for the classification of SAR images, including difference variance、difference average、difference entropy、contrast、energy、variance、sum variance、inverse difference moment and correlation etc. Furthermore we have abstracted features of SAR image and studied classification characteristic. Using criterion called distance of inside classes and between classes, we can get a few features of a SAR image which is good at image classification. At last, making use of BP neural network of three layers, we proceed image classification and get the satisfied results.