Abstract: A new image clustering segmentation method is proposed by the partition of rough tolerant relations. First, using FCM clustering results with different cluster numbers, an information table can be constructed and the image will be divided into many object regions. Secondly, by applying ‘value reduct’ , the weights of the attributes can be acquired and the dissimilarities between objects are calculated. Then tolerant relations can be defined based on dissimilarities between regions. Finally, tolerant relations partition the objects and accomplish the segmentation of the image. The method was applied to an artificial generated image and a human brain MRI image. The results of the experiment indicate that the method proposed here has better segmentation veracity than FCM method and has well performance for vague boundary region segmentation.