The current proposed methods for fault diagnosis of power systems， including expert system， artificial neural network， optimization method， Petri network， rough set theory， fuzzy set theory and multi-agent technology， are surveyed in this paper. The characteristics and applicability of those methods is briefly analyzed and also， the existing drawbacks and their improvements are given. Finally， several key technical issues and the main development trends of the topic are pointed out to solve the practical problems and motivate further developments.