Twelve typical faults and fuzzy treatment of nine symptom parameters were analyzed. A fault diagnosis method for a fuzzy Kohonen neural network was proposed based on diagnostic working principles and specific features of a Kohonen neural network. The application of the method shows the following merits: a selflearning function, rapid operating speed, and strong grouping capability. The fuzzy Kohonen neural network can diagnose single and multiple faults. It is an effective and suitable method for fault diagnosis of the regenerative heating system of the steam turbine unit.