Due to the variety and the complexity of marrow cells image, better results may be achieved by adopting multi-threshold for the image segmentation. However, with regard to no significant peak and valley values in the histogram, it is too difficult to determine thresholds. Therefore, a new method of combining the spatial scales filter and the wavelet transform analysis is adopted to find suitable thresholds and the obtained edge is optimized by the active contour model. Proposed a new genetic clustering algorithm based on entropy to classify the marrow cells. Through the experiences, better result can be attained.