Abstract: Event correlation is a key problem of network management. Though quite a number of event correlation methods have been proposed to solve this NP complete problem, none of them can deal with it completely. A hierarchical event correlation technique based on intelligent agents is proposed in this paper. It can be divided into two parts: ① The intelligent agent, which resides on a network device, classifies, filters, and correlates all the alarms emitted by the device directly; ② The correlation results of all intelligent agents are collected by an event correlation server that resides on a network management center. According to different kinds of data, knowledge, and experience, this server application uses model based reasoning and case based reasoning techniques to reason out the most possible fault hypothesis. This method decomposes the complexity of event correlation problem, so it is of great benefit to realizing real time and automated network management. Additionally, the event correlation system in this method is general enough to be used in different networks, and is adaptable to the evolvement of a network.