Abstract: Small water body extraction of hilly area from TM images is an important and difficult task. For the large differences in the characteristics of brightness value of water body and shadow in blue-green bands and typical model of multi-band spectal relationship, this paper build a new water information ertraction model of multi-band spectal relationship on the basis of the difference operation. Firstly, water bodies are extracted from shadow and other surface features. Secondly, to connect short-terms and fill empty based on mathematical morphology filtering algorithm. Finally, adoption of binary image thinning algorithm, objective water bodies are refined. After experimental analysis and comparison show that the method overcomes the shortcoming of lot of water extraction methods wihich can only effectively extract larger water bodies. The means not only can automaticly and high-precision extract small water of hills and mountains area, but also can effectively remove the shadows and other interference information.