With the increasing presence and adoption of web services, accurate QoS prediction methods are becoming increasingly important. Although some QoS prediction techniques have been proposed and analyzed recently, the performance is not satisfactory, since they didn’t take the relation between QoS values and users’ physical locations into consideration. In order to improve the precision of web service QoS prediction, we propose a probabilistic matrix factor model that fuses the users' own properties and their physical neighbors' performance together, so as to make a comprehensive use of both the users’ QoS records and their location relation. The experimental results show that our method performs better than the state-of-the-art approaches.