The commonly used cluster based segmentation method assumes that the sample distribution is hyperspherical, but this kind of assumption is not consistent with the real characteristic of the human brain MR (magnetic resonance) image. In order to surmount this drawback, a new algorithm for segmenting MR image based on hyperellipsoidal fuzzy clustering is presented in this paper. Provided experimental results indicate that the proposed strategy is feasible for classifying the white matter and the gray matter of the brain, and has the merits of both high efficiency and remarkable accuracy.