Network intrusion detection is one of the hottest research issues on the network security recently. In the paper, three kinds intrusion detection model are introduced. By using Bayesian methods, Fuzzy Decision Engine and Immunity-Based Model, we can improve the performance of intrusion detection system and effectively reduce the false positive alarm. Based on the three models, some drawbacks of NIDS have been analyzed and a new architecture of IDS has been put forward. In this new architecture, IDS are viewed as a function instead of a product.