The open source software (OSS) development communities have experienced rapid growth in recent years. Previous social network studies on OSS communities focused on collaboration relationships. However, information about how OSS community members perceive each other is largely ignored. In this study, we report an empirical investigation of the evaluation network in an online OSS community which includes over 11,800 OSS projects and more than 94,330 developers. A collaboration network is modeled from this data set and analyzed for comparison purposes. We find the evaluation network is significantly different from collaboration network in average degree, average path length and fragmentation rate. Furthermore, we argue that the evaluation networks can be used to locate expertise - skillful developers in OSS communities and capture important social relationships among the developers missed in the collaboration network. These characteristics of the evaluation network may benefit the research of OSS development communities and expert recommendation systems.