This paper is concerned with the problem of detecting and isolating multiple faults by a special structure of the full-order Kalman filter. A new state filtering strategy is developed to detect and isolate multiple faults appearing simultaneously or sequentially in discrete time stochastic systems. Under a fault isolation condition, the proposed method can isolate q simultaneous faults with at least q output measurements. The fault isolation filter generates a reduced output residual vector of dimension q so that its ith component is decoupled from all but the ith fault and so that the effect of plant and state noises is minimized. Necessary and sufficient conditions for stability and convergence of the proposed filter are established.