The design of an adaptive computer-assisted instruction (CAI) system which aims at efficient and effective training of knowledge and skill necessary for diagnosing nuclear power plant anomalies is described. The CAI system has the following functional features, (1) On-line measurements of human information activities; eye gaze point and think aloud, (2) Estimation of human internal states; thinking process, and (3) Adaptive interaction to individual trainee in accordance with the estimated internal states. And, the new elemental technology necessary for realizing such CAI system, which is the real-time eye gaze point detection using Eye-Sensing Head-Mounted Display, is explained. Then, the design of an experimental CAI system which aims at acquiring the knowledge of effective instruction by conducting a laboratory experiment, and a training process using this CAI system are described. In this system, a human expert, who is a substitute for instruction by computer, will provide some instruction to a trainee.