Blackouts which happened continually these years around the world show that power system desiderates more effective means of security monitoring and fault diagnosis. A new online fault intelligent diagnosis approach is proposed in this paper， which is based on distributed characteristics of power flow before and after fault. This approach can track operation in an adaptive way， select monitored data and monitored objects dynamically and pre-create the library of fault patterns through analyzing the relationship between distributed characteristics of power flow and the outage of electric power network. While distributed characteristics of power flow matches a pattern in the library， the approach realizes online intelligent diagnosis of the fault. The results of numerical tests have justified that the approach is intelligent， accurate and robust. It can enhance operator's ability for grasping gird state of affairs and dealing with grid accidents.