A component-wise expert system (EXRCP) has been developed to diagnose reactor coolant pump (RCP) failure and to offer corrective operational guides to plant operators. This system consists of three main parts : an inferencing mechanism, a knowledge base and a user interface. The system performs alarm processing by rule matching and priority factor operation. From the alarm processing result, the system can diagnose the primary causal alarm. To find the cause of RCP failure, the system performs rule-based deduction and Bayesian inference. The system first tries to find a definite cause by the rule-based deduction. If this fails, the system solves probable causes by Bayesian inference. When the diagnosis is finished, the system enables the operator to take corrective actions. The knowledge base is partitioned into alarm basis knowledge units. In this way, the system is concerned only about the knowledge unit of the primary causal alarm. Moreover, the execution time of the system is considerably saved.