[en] A failure cause analysis method to be used in a failure diagnosis system for large tokamak fusion machines is proposed. The diagnosis system is designed to identify causes of the abnormal states indicated by alarm annunciators in the machine. The proposed method is based on a cause consequence tree (CCT) model. The analysis comprises two steps; (1) selection of several possible causes from all causes in the CCT, (2) search for the most probable cause from among them. Simulation tests on the circulation system for cooling of a tokamak machine show that this method correctly selects causes in rate of more than 90%. Average computation time for cause determination is within 17 seconds. (author)