The increasing P2P network traffic on the Internet has leaded to the problem of network congestion. In the consequence of the diversification of the P2P business and protocol, research on the management of P2P traffic has many problems to resolve. Prediction of the P2P traffic is the kernel problem in the P2P traffic management. Based on the existed P2P traffic characters, this paper structures a P2P traffic model, gives a traffic prediction algorithm bases on wavelet-analysis, and proves the accuracy of the algorithm. Simulation experiment figures that the algorithm has a high prediction precision and a superior real-time performance.