Two kinds of identification techniques for the diagnosis of disturbances in nuclear power plants have been proposed, and the applicability of these techniques to actual plants has been verified by computer experiments. In both techniques, a set of the observed signals (observed vector) obtained from an actual plant is identified with one of the categories representing a normal state, several anticipated anomalous situations, and an unanticipated anomalous state, in which the categories corresponding to the anticipated anomalous situations are classified by the kind and approximate magnitude of the anomaly source (the disturbance). The maximum likelihood technique is used in method 1. It applies to the identification of multiple anticipated disturbances that happen sequentially with some time interval, even if a plant has some nonlinear characteristics. The projective operator technique is used in method 2. It applies to the identification of any kind of multiple anticipated disturbances under the conditions of the plant having approximately linear characteristics and the observed vectors corresponding to the anticipated disturbances are linearly independent of each other.