Aiming at two disadvantages of the traditional HCM, which are its too long running time and its easy getting into local optimum solution, the HCM algorithm combined with pyramidal structure and subtractive clustering is put forward. This algorithm uses the pyramidal structure to describe the image in different scales, and carries out HCM after getting the initial center by the subtractive clustering in the top image. Then HCM is done in the image at this layer with cluster result of upper image as the initial center, the result of the bottom image is the final clustering result. Simulation experiments show that the running time of this algorithm is far shorter than that of traditional HCM, and clustering quality also is better.