This paper proposes a computational grid platform for solving the large-scale power system applications. The platform is based on Globus Toolkit middleware and GridWay meta-scheduler. It can enable large-scale sharing of computational resources across institutional boundary. This paper first discusses the architecture and each component of the platform, and then the test bed is described. Finally, the test results of probabilistic load flow (PLF) by Monte-Carlo simulation are presented. The test results show that the computational Grid system can provide comparable performance.