By using the three-way decision ideas in rough sets, the class is characterized by positive domain, negative domain and boundary domain to get the initial clustering results. The final clustering result was merged by defining the degree of overlap and the merging strategy of class. The two given relational network data have shown the specific clustering steps, and by comparing the clustering results of two examples, a major factor influencing the clustering results is arisen: the selection of threshold value. Experiments show that: the effect of threshold value selection on clustering results of simple network structure datasets is not obvious, but the impact of clustering results on complex network structure datasets is more significant.