Abstract: One mission of network management is to monitor the network effectively. The monitoring methods in Simple Networks Management Protocol (SNMP) are polling and trap. Both of them consume the system resources and network bandwidth. Based on the hierarchical network architecture, a dynamic monitoring algorithm is proposed so as to minimize the management overhead. By analyzing the probability of the alarm occurrence with the sampled network information, the proposed algorithm prolongs the interval between two consecutive polling when no alarm could occur. It also reduces the redundant alarm information resulted from the signal fluctuation. This method significantly reduces the amount of monitoring traffic and guarantees to detect the abnormal event. The simulation results show that the proposed monitoring scheme reduces both the number of polling and the cost of the network management, consequently improves the network performance.