The traditional way of dividing the power grid is mainly based on geographical and administrative area, which often ignores the network’s structure and characteristics, however, power grid’s blackout always results from the partition lines. Community structure detection of complex network is used to identify the network’s topology partition by the connections between nodes and it has a familiar structure with power grid’s partition. This paper introduces the conceptions and evaluation indexs of community structure and the modularity Q function, comes up with a new power grid partition method based on community structure detecting. Four different kinds of weighted network models are built with grid tide and impedance data. The improved Louvain hierarchical algorithm is adopted to divide power grid and a practical improvement is made in partition process considering the real power grid’s characteristics. Through the community division results of IEEE 39, 118, 300 standard power grids and China southern power grid in different models, this method is proved to be effective and accurate.