Medium and long term load forecast is an important base of transmission and distribution network planning. Due to the effect of some relative factors on the load, raising the precision of load forecast will be a challenging task. Based on the theory of rough sets, the relationships within the load data and relative factors are revealed, and the most sensitive factor can be found. By clustering, the model of load and most sensitive factor can be built and used to load forecast. The validity and effectiveness of the method is tested in a real power system example.