To segment fuzzy and touching cell images accurately, an image segmentation algorithm based on graph theory and morphological mathematics was proposed according to the characteristics of medical cell images. With proposed algorithm, the images were smoothed and sharpened, firstly. Then, the improved Minimum Spanning Tree (MST) algorithm was used to segment the cell images, in which the cell size and shape information were added into MST graph to avoid the over-segmentation. Furthermore, the adherent cells were split by combining the distance mapping and the skeleton information in morphological mathematics to solve the problem of cell adhesion in the binary image. Different from the traditional watershed algorithms, the split algorithm has no repeat operation. By experiments, it shows that the proposed algorithm can segment fuzzy and touching cell images well and can obtain desired results.