A real-time analysis method is proposed to trace cognitive process in diagnosing plant anomalies. The method under development is to apply for a real-time cognitive process estimator, with the aim of realizing adaptive interface for training the skill of plant anomaly diagnosis. The major point of the real-time estimator is that it is equipped with the Causal Relation Diagram (CRD) which represents the causal relationship between the major nuclear power plant (NPP) parameters. By using CRD and two types of information, one from data of operational sequence to user interface, and the other from verbal information taken by speech recognition machine, this estimator computes degree of operators¡Ç confidence in judging primary cause of plant anomaly for each major NPP parameters in CRD. The method, function of the estimator and the result of experimental evaluation are described in this paper. (in Japanese)