This paper proposed a new power grid fault diagnosis model combining Petri nets and grid computing. It includes the power grid computing system's structure, software service and algorithm. The model can use the high performance computing method and distributed environment of grid computing. At the same time, Data grid system can offer table, efficient and unified information resources for the power grid fault diagnosis. When a fault happens, the component connecting topology trees were used for creating the Petri Nets model of possible fault components. Finally according Petri Nets model the true fault primary devices were identified, and the false tripping or operating information were picked out. This fault diagnosis model standardized the fault information and can share fault diagnosis results with other systems, realized real-time online fault diagnosis of power systems. (4 pages)