In this paper, a novel method is proposed for solving an open problem of shape matching in content-based image retrieval. Our method regards the stack-triangles geometrics histogram to be the feature of an image, the correlation of two images was decided by the compare result of their stack-triangles geometrics histograms. First of all, these methods extract the key contour fragments of an image, so that a shape can be presented as many key contour fragments in this way. In addition, we derivation stack triangles of a shape by the key contour fragments we get it before. Then, we can get the stack-triangles geometrics histogram of the image and calculate the correlation of two images use their stack-triangles geometrics histogram. Finally, experiment results show that our new method is a very fast method and can get more robustness than existing methods and achieve a good effect in the standard shape databases MPEG-7.