This paper is concerned with an active fault diagnosis method is developed based on an adaptive fault detection approach using χ 2-statistics [1]. The system under consideration is linear discrete time-variant with sensor faults in the framework of mixed uncertainties (stochastic noises and bounded uncertainty for parameter matrices). The proposed method enhances the fault detection performance and provides the ability of localization and estimation of the detected fault. It provides also an on-line fault diagnosis with no delay in time instant with the computation time depending only on the computer performance.