A method for nuclear power plant accident diagnostics using optimal control theory was investigated. A computer code, CAAD (Computer Assisted Accident Diagnostics), based on Pontryagin’s Principle, has been written and applied to single control malfunctions in nuclear power plants. Using a simplified systems model to represent power plant dynamics, the control theory in CAAD is expected to be suitable for expanding to more detailed systems applications. Results were obtained for a series of simulated plant transients that compare key state parameters to the control parameter output resulting from searches for the correct malfunction. A performance index is used as a convergence criterion upon which to define the degree of approach to the correct failure identification. The long-term goal of this methodology would be to expand to on-line computer application in reactor control rooms. Intended to be automatically initiated by a reactor accident, the malfunction search process would provide the operator with additional information to assist in identifying the most likely failure.