A comparison of the effectiveness of three algorithms for fault detection and isolation of SINS sensors when using triple redundancy is made. To increase the efficiency of solving the problem, the authors proposed three modifications of the pairwise difference method. Methods which use weighted sum of pairwise differences, calculate an adaptive response threshold, and transform to principal components are proposed. In most existing approaches, fault detection is done at the system level, namely, the values of the output navigation and dynamic parameters of the SINS are used. In contrast, we propose to solve diagnostic problems based on the inertial sensors’ measurements, which reduces the time elapsed from the occurrence of a fault to its detection. In addition, solving fault detection problems at the level of inertial sensors opens opportunities for reconfiguring the SINS to ensure the accuracy of the redundant system in the event of fault of one or more sensors in its composition. Comparison of algorithms is carried out through semi[1]natural simulation, in which artificial faults are added to the data obtained under conditions of a stationary base. Additive single and repeated faults with random amplitude are considered as an example. Relative to the values measured by the sensors, faults amplitude can be classified as small, medium and large. To compare the effectiveness of the proposed algorithms, a confusion matrix is calculated, from which the precision and recall metrics are then calculated.