Abstract: In view of the existing algorithm that is unable to take better account of the network connectivity and the attributes of individual nodes comprehensively, the limitation of the typical algorithms of agglomerative and divisive clustering was analyzed, thus defining conceptually the edge loading, edge weight, connectivity threshold and graph segmentation. Then, a new algorithm SoNetCD based on BFS(breadth-first-search) is presented for discovering the communities in social networks, which takes both network topology and edge weight into consideration. Inter-community edges are cancelled to reveal the community structure in the algorithm, thus judging exactly the inter-community and lowering the failure rate of cancelling inner-community edges. Experimental results of a real-world social network dataset showed that the SoNetCD outperforms the typical GN algorithm in identifying community structure.