In order to solve the problem that parameters are difficult to obtain in Bayesian network（BN）， the subjective Bayesian approach is introduced into the process of fault diagnosis of a power system. The BN is rationally simplified before inferring with the subjective Bayesian approach， which makes the approach more efficient. Two more approaches， i.e. inference with uncertainty reasoning and comparison of the number of abnormal events， are proposed to deal with the uncertain and lost information. The two approaches can be used in conjunction， however， their computing speed will be too low when too much information gets lost. A group of rules are thus adopted to infer the most likely state of lost information and ignore the unimportant information so as to improve the computing efficiency. Finally， the results of several tests show the scheme proposed is rational and practical.