Abstract: Stepwise clustering, one of the methods for non-hierarchical classification, has been introduced in this paper, and was applied to the classification of Elaeagnus mollis community in Shanxi. The results show that the stepwise clustering accomplished the objective process of the optimal classification through minimizing the sum of the squared deviations within plot group and by maximizing the sum of the squared deviations between plot groups. This led to minimum homogeneity within plot group and maximum heterogeneity between plot groups. The results of stepwise clustering tallies with the reality. Furthermore, it allows the work more efficient because we only need calculate the centroid distance from one sample to another. Compared with fuzzy c-means algorithm and with TWINSPAN, the result of stepwise clustering is similar to that of fuzzy c-means algorithm which has greater homogeneity within plot group. In addition, the stepwise clustering is superior to the TWINSPAN procedure, providing that the classification results do not require an obvious hierarchy among plot groups.