The growing ubiquity of mobile devices allows users to form mobile social networks (MSN) which create new means for virtual interactions on the basis of their users’ physical features.Many MSNs can benefit from the knowledge of their users’ real-world social relationships.In this paper,we propose an approach for estimating users’ real-world interpersonal social relationships by leveraging two kinds of physical features,i.e.location and proximity which are respectively represented by GPS locations and Bluetooth detections in mobile environment.To estimate the real-world social relationships between a user and all his encounters,the approach firstly extracts places with semantic meanings (e.g.home,work,etc.) where the user has visited,and then the social relationships are estimated by integrating both the extracted places and the proximity data.The experimental results show that our approach can achieve an accuracy of over 90% in estimating three types of real-world social relationships (i.e.family,colleague and friend).