In order to segment fuzzy and touching cell images accurately, an improved algorithm is proposed based on minimum spanning tree (MST) and concave point detection. First, the cell images are smoothed and enhanced by a Gaussian filter. Then, the improved minimum spanning tree algorithm is used to segment the cell images. The MST algorithm is modified from three aspects, namely, weight function of edges, difference function of internal and inter region, and threshold function and parameter k. Furthermore, the problem of cell touching is solved by means of concave point detection. According to the rugged topography of touching cells, the concave points are found from the concave regions in the touching cell images, which are used to find the separation points quickly and accurately. Experimental results indicate that the new algorithm is ideal and effective.