An approach to predict failure in the measurement system is developed. Proposed approach uses confidence and tolerance intervals for the innovation of Kalman filter and it is based on the criterion of overlapping these intervals whereby it allows the detection of potential failures in the system. The proposed failure prediction method is well suited to realize detection of constant shifting and slowly varied failures, such as distortion in measuring instruments and gyroscope axes shifting in navigation systems. The method is simple and permits to solve the problem of failure prediction in the measurement system under the conditions of its practical operation. (C) 2011 Elsevier Ltd. All rights reserved.