A classified computer network is vulnerable to external and internal attacks in the internet environment. It is difficult to get a lot of training sample of the security computer network system at the actual situation. This paper aimed to analyze the security of the computer network under the distributed intrusion detection. The authors studied how to increase the intrusion detection accuracy in the case of small samples, so that to keep and deal with the invasion of the network timely. In this paper, they proposed a new intrusion detection method based on improved SVM Co-training. The specific implementation process of the algorithm is presented.