We used a cluster-based method to investigate the diffusion-attenuated signal of glioma patients with different grades. The clustering results were analyzed by diffusion spectrum, which returns a continuous distribution of diffusion coefficient for a given attenuated signal. CSF, gray matter and white matter were clearly separated by Fuzzy C-means clustering. And some clusters showed sensitivity to interface between glioma-related tissues and normal tissues, which can be used for tumor delineation. High grade glioma tended to have clusters with smaller diffusivity and contained more types of clusters than low grade glioma.