We have developed a method to estimate a plant operator's thinking state during abnormal situations. The operator's thinking state is estimated from EEG data. First, three basic modes were defined according to the typical thinking slate of a plant operator at the time of a malfunction. Second, experiments were performed to evoke the basic modes. Third, a nonlinear model was developed to estimate the thinking state of a plant operator. Finally, the estimation method was applied to a subject who operated a boiler plant simulator under abnormal circumstances. It was verified that the method gives useful information for estimating the plant operator's thinking state.