A novel technique is introduced to detect and isolate the failures of multiple actuators connected to a system. The failure of actuator considered in this study could be any type of erroneous input that is different from the commanded one. The interaction matrix technique allows the development of input-output equations that are only influenced by one target input. These input-output equations serve as an effective tool to monitor the integrity of each actuator regardless of the status of the other actuators. Although the procedure requires the knowledge of analytical model of the system being tested, the analytical redundancy can be experimentally predetermined through standard input-output-based system identification algorithms such as observer/Kalman-filter identification (OKID) and eigensystem realization algorithm (ERA). This method is capable of real-time actuator failure detection and isolation under any type of input excitation. Both numerical simulations of a spring-mass-damper system and a laboratory experiment using eight-bay NASA truss structure verify the feasibility of the proposed method.