In order to detect, identity and hold up network attacks, a network intrusion detection system based on rough set theory and multiclass linear support vector machine (linear SVM) is in this article. The system makes the most of rough set theory and linear SVM to reduce the redundancies of data sets and improve the detection rate of EDS. The simulation experiment shows this approach has higher ratio of correct classification, while shortens training time of the classifier in a wide range, which is going to a pretty momentous improvement in real-time detection. On the other hand, this approach has reduced memory usage and improved the generalization ability of the system.