A sensor fault detection method combining the single sensor parity relation (SSPR) with the likelihood ratio test (LRT) is described. The SSPR is in an algebraic form that correlates system dynamics with multistep readings of a sensor and is therefore fast running. The scheme can easily be duplicated for each sensor of interest and thus has advantages for modular design and parallel processing. In the fault detection architecture, residuals generated from the SSPR module are examined by an LRT module for failure signatures. The likelihood ratios are maximized according to the fault occurrence time to improve detection sensitivity and are then calculated using a recursive form to match the speed of the SSPR module. The proposed concept is demonstrated with hypothetical sensor failures for pressurizer instruments. Comparisons with the Kalman filtering technique and the sequential probability ratio test are discussed.