With the fast-growth of mobile social network, people’s interactions are frequently marked with location information, such as longitude and latitude of visited base station. This boom of data has led to considerable interest in research fields such as user behavior mining, trajectory discovery and social demographics. However, there is little research on community discovery in mobile social networks, and this is the problem this work tackles with. In this work, we take advantage of one simple property that people in different locations often belong to different social circles in order to discover communities in these networks. Based on this property, which we referred to as Location-Interaction Disparity (LID), we proposed a state network and then define a quality function evaluating community detection results. We also propose a hybrid communitydetection algorithm using LID for discovering location-based communities effectively and efficiently. Experiments on synthesis networks show that this algorithm can run effectively in time and discover communities with high precision. In realworld networks, the method reveals people’s different social circles in different places with high efficiency.