The skeleton is an effective representation of object shapes. Methods based on distance transformation can generate skeletons with accurate position, but connectivity of the skeleton branches must be carefully checked. It is difficult to check the connectivity when the structures of the skeleton are very complicated. In this paper, a dynamic skeleton algorithm for binary images is presented by combining with thinning method and Snake model technique. First, compute the initial skeleton based on thinning technology in which the connectivity and the topology of the skeleton can be kept well. Then the initial skeleton can be led to its accurate locations in distance field based on Snake model technique. The skeletons generated by this method are not only locating at accurate positions, but also with correct connectivity and topology.