In multi-user networked virtual environments such as Second Life, 3D streaming techniques have been used to progressively download and render 3D objects and terrain, so that a full download or prior installation is not necessary. As existing client-server architectures may not scale easily, 3D streaming based on peer-to-peer (P2P) delivery is recently proposed to allow users to acquire 3D content from other users instead of the server. However, discovering the peers who possess relevant data and have enough bandwidth to answer data requests is non-trivial. A naive query-response approach thus may be inefficient and could incur unnecessary latency and message overhead. In this paper, we propose a peer selection strategy for P2P-based 3D streaming, where peers exchange information on content availability incrementally with neighbors. Requestors can thus discover suppliers quickly and avoid time-consuming queries. A multi-level area of interest (AOI) request is also adopted to avoid request contention due to concentrated requests. Simulation results show that our strategies achieve better system scalability and streaming performance than a naive query-response approach.