Peer-to-Peer technology is one of the most popular techniques nowadays, and it brings some security issues, so the recognition and management of P2P applications on the internet is becoming much more important. The selection of protocol features is significant to the problem of P2P traffic identification. To overcome the shortcomings of current methods, a new P2P traffic identification algorithm is proposed in this paper. First of all, a detailed statistics of traffic flows on internet is calculated. Secondly, the best feature subset is chosen by binary particle swarm optimization. Finally, every feature in the subset is given a proper weight. In this paper, TCP flows and UDP flows each have a respective feature space, for this is advantageous to traffic identification. The experimental results show that this algorithm could choose the best feature subset effectively, and the identification accuracy is improved by the method of feature weighting.