Abstract: With the development of power system,the data in power system is increasing rapidly. Because of the diversity and complexity of the data,it has Multi levels and randomness,which brings the problem of data mining. This paper presents a theoretical method for data mining based on rough set,by rough sets for take the minimum attribute reduction set,scanning the decision table reduction form,to distinguish the key signals and non critical,to find out the diagnostic rules from the sample set,achieve rapid fault diagnosis. Finally,examples prove correctness of the algorithm in power system data mining.