In order to improve the protection level of network security, a combined anomaly intrusion detection method based on the partial least squares(PLS) method and the kernel vector machine(CVM) is proposed. The PLS algorithm is used to extract the principal components of the network data and construct the feature set. Then the CVM is applied to build the anomaly intrusion detection model of the feature set, completing the detection and decision of abnormal intrusion. Simulation results show that the proposed method has the fast processing ability similar with the CVM for the large scale data, and the detection performance is roughly equivalent to that of L1-SVM and L2-SVM, and the detection level is relatively high when the principal component is 1538, which proves the effectivenes and feasibility of the method in the application of anomaly intrusion detection.