Automatic control of routine plant operation is receiving increasing attention as a valuable tool for improving plant performance. A crucial aspect of automatic control is the capability to manage malfunctions. Among the tasks involved is the isolation (identification) of the malfunctioning apparatus. An algorithm for malfunction isolation in linear stochastic systems is developed. It is shown that a single linear filter is adequate for isolating a wide range of malfunctions. Most importantly, no knowledge about the nature of the malfunction is required to construct the filter, other than that the linearity of the dynamics and the measurements be preserved (complete or “hard” sensor failures are included). It is shown that the performance of the algorithm improves with the number of state variables that are directly measured. Numerical application to a simple nuclear plant model is presented.