Abstract: In the context that social network becomes more and more complicated and huge, it is extremely difficult and complex to mine the global community structures of large networks. Therefore, the local community detection has important application significance for studying and understanding the community structure of complex networks. The existing algorithms often have some defects, such as low accuracy and stability, the preset thresholds difficult to obtain, etc.. In this paper, a local community detecting algorithm is proposed based on boundary nodes identification, and a comprehensive consideration of the external and internal link similarity of neighborhood nodes for community clustering is given. Meanwhile, the method can effectively control the scale and scope of the local community based on the boundary node identification, so as the complete structure information of the local community is obtained. Through the experiments on both computer-generated and real-world networks, the results show that the proposed algorithm can automatically mine local community structure from the given node without predefined parameters, and improve the performance of local community detection in stability and accuracy.