Abstract Based on a fuzzy C-means (FCM) clustering algorithm, the height data set of a measured grinding wheel surface is classified into two fuzzy clusters, which are named as “grain” and “bond” respectively. The clusters centers are initialized with the centroids of the classified data first, and then the most optimal cluster centers and the membership matrix are obtained with an iterative strategy. By choosing appropriate thresholds of the membership degree and the Euclidean norm of the two cluster centers, the edge of grains is determined. The method is validated with experiment. For further analysis, an evaluation is carried out by applying this method to a simulated grinding wheel surface, the results of which show that the error of this method is less than 2.0%.