This paper studies the performance assessment and regulation for power enterprise, and proposes a grid performance assessment method based on Data Envelopment Analysis (DEA) considering the power grid operation characteristics, the performance assessment requirements, and the data availability. In the method, R-clustering technology is used for input and output selection. Based on the input and output index, Q-clustering technology is used to classify the power grids. And then the validity of the classification is tested by statistical criterions. Furthermore, the power grids belonging to the same cluster are treated by DEA. The grids which can not be analyzed effectively by DEA are projected by introducing the slack variables to illustrate the optimal route for the performance improvement, which is a delicacy DEA performance method. In the end, real data of the 11th-Five-Year Plan period and 2011 provided by China Electricity Council are used to conduct an empirical analysis of provincial-level power grid performance and it illustrates the effectiveness and practicability of the proposed method.