The article presents the problem of the sensor fault diagnosis in the unmanned underwater vehicle control systems described by means of the structure schemes including blocks with transfer functions and static nonlinearities. One of the most important components of the control system of the unmanned underwater vehicles is navigational sensors. The problem of a timely detection of faults in this group of sensors is very important. Faults in the sensors may cause incorrect execution of the task or even loss of a vehicle. The sensor fault detection and isolation in the unmanned underwater vehicles control system is based on the concept of an analytical redundancy. Analytical redundancy includes two or more ways to determine the values of the variables of the system, one of which uses a mathematical model, presented in an analytical form. One of the methods of the system diagnosis based on analytical redundancy is the observer-based approach. Diagnostic observers are based on a mathematical model of the diagnosed object. The decision is based on the analysis of the residuals generated as a result of mismatch between the outputs of the sensors and the outputs of the observers. In a healthy system the residuals are close to zero. When a fault occurs, the residuals become significantly different from zero. The problem of interest to us is development of a fault isolation observer-based procedure with an accuracy of a block of the initial system (if this is possible). Let us assume that only input and output signals are available for diagnosis. The reason for these restrictions is that a model of the initial system does not go through any nontrivial linear transformations. Therefore, this approach can be used in a nonlinear case.