The concept and development of a fault-tolerant control system and the results of its real-time application using commercial microprocessors are presented. The system incorporates (via recursive filtering) on-line detection and reconfiguration of faulty equipment, sensor calibration, and information display within a structure that relies on alternatively configured regulators for plant control under different operational modes; these regulators are designed on the basis of analytically derived control laws and using a rule-based heuristic approach. Each critical input signal is a weighted average of all valid measurements of the appropriate plant variable where the weighting matrix is adaptively updated as a function of a posteriori probabilities of failure of these redundant measurements. Since the weight of a degraded measurement is smoothly reduced, eventual isolation of this fault does not cause an abrupt change in the estimate of the measured variable, and therefore, the feedback control system remains ``bumpless.'' The automated system has been shown to be tolerant of equipment failure(s), sensor degradation and noise, and certain types of process malfunctions, disturbances, and uncertainties by experimentation at the MIT nuclear research reactor.