Abstract: To overcome the difficulty caused by the overlapping of cells in the image analysis for cancer diagnosis, a new cell image analysis method is proposed based on watershed algorithm to separate the clustering cells into single ones. Firstly, a distance map is produced to determine the number of seeds from the clustering cells based on iterative erosion. Then, the ultra-erosion aggregates are taken as the seeds to separate the cells. Finally, the images are grown from the seed regions until all of the image pixels except the boundaries of convex have been assimilated. The design and implementation of the algorithm and experimental results are presented.