The choice of super nodes is an important factor affecting the fluency and playback quality of P2P streaming media systems. Aiming at the characteristics of P2P system, a selection method combining node reputation and service capability is proposed. First, according to the interaction behavior and service evaluation between nodes in the network, the reputation value of each node is calculated by Bayesian theory. Then, the service capability parameters are obtained according to the computing power, storage capacity, memory size, and bandwidth of each node. When there is a service request, the physical distance between the requesting node and each node is estimated by the IP address. Consider the above factors to select the last super node to serve the request. Simulation results show that this method can effectively improve network throughput and service performance.