With the massive increasing in the number of users on the Internet, a huge burden has been placed on the network servers and various internet companies. Peer to Peer (P2P) network breaks the traditional server operation mode, which can reduce the use of servers and other network overheads and has been used by more and more network administrators. Aiming at the defects of the current P2P network traffic identification methods, such as large error and poor identification effect, a method of identifying P2P network traffic by support vector machine (SVM) network is proposed. Firstly, we preprocess the public dataset provided by Andrew Moore to establish the training set and test set composed of feature vectors of P2P traffic; then we use SVM network to train the training set and establish the recognition model; finally, we use MATLAB to carry out simulation experiments and compare the results with back propagation (BP) neural network and radial basis function (RBF) neutral networks, which show that the recognition accuracy of P2P traffic based on SVM network is higher and can be used efficiently to identify P2P traffic. Network traffic recognition accuracy can reach 95.07%, recognition results tend to be stable. The P2P traffic identification method based on SVM network can manage the P2P network and has high practical value.