The parameters setting and deflective of classification hyperplane caused by imbalance data affect the SVM’s application in network intrusion detection. To resolve these problems,a new method based on glowworm swarm optimization SVM(GSO-SVM)was proposed. The algorithm optimizes SVM training parameters,and a modification gene was proposed to modify the classification hyperplane as well. The glowworm swarm optimization method was introduced to this optimization problem because of its excellent ability of escape locally optimal solution. The intrusion detection rate reaches 97.33%,higher than SVM and SVDD method. The experiments results show that GSO-SVM can improve the SVM’s generalization and get lower false alarm rate and failed reporting rate.