Three fault diagnosis methods,i.e.fuzzy inference,rule-based inference and the combination of both,are analyzed and their shortcomings in studying the relation between symptoms and causes of mechanical fault are also pointed out in this paper.Based on these,an advanced method is educed,in which the fuzzy neural network(FNN)instead of fuzzy inference is used and combined with rule-based inference.The result of a sample diagnosis on a real car shows that the method is effective.