The peer-to-peer paradigm is promising for large-scale point-to-multipoint multimedia streaming. When a peer receives a media stream, it relays this stream to other peers (i.e., it acts as their parent). When new peers join or existing peers leave the system, it is necessary to select parents for the new or the affected peers. There are two existing approaches to parent selection: i) static approach selects parents based on peers' location information which is inferred from the peers' IP addresses, and ii) dynamic approach selects parents based on the dynamically measured network conditions. The static approach does not have measurement overhead but may not accurately select good parents, while the dynamic approach can select good parents but has significant measurement overhead. In this paper, we propose a hybrid approach to parent selection. This approach integrates the static and dynamic approaches to obtain their respective advantages. It uses static information (namely, registration information of IP addresses) to identify the geographically near peers, and then uses dynamic measurement to select one of them as the parent. We demonstrate that the hybrid approach can select good parents while it has significantly smaller measurement overhead than the dynamic approach.