A fault diagnosis system incorporating object-oriented programming models into a neural network is developed and reported in the paper. At the same time, to draw an inference efficiently, back-propagation learning rules, statistical process control, and alpha-beta depth-first algorithm are also embedded in the system. For the purpose of fault diagnosis, the object-oriented multilayer perceptron network is first trained by the back-propagation learning rule. Then, the statistical process control is used to analyze the trends by historical data and detect suspicious components. At last, by means of the alpha-beta search technology, the most plausible fault candidates and the rank of those candidates are generated speedily.