All of the former level set algorithms have only one level set function and only one speed function, and it is a complex procedure to minimize the energy function during the evolvement of the zero-level-set. Furthermore, there are a lot of problems in this single speed function. In this paper, a new multi-level-set algorithm with multiplicate speed functions is proposed according to the different properties of different objects: Different level set functions are constructed in different regions, and so are different speed functions accordingly; many zero-level-sets are evolved at the same time and act on one another in order to segment. This method not only enhances the accuracy of segmentation, but also solves the bounder gap problem well, which is quite a puzzle for single level set algorithm. Perfect results are achieved when this method is applied to segment the MR and CT images.