A novel sensor failure detection method is developed in this paper. Sensor failure considered in this paper can be any type of measurement error that is different from the true structural response. The sensors are divided into two groups, sensors that correctly measure the structural responses, are termed as reference sensors, and sensors that may fail to correctly measure the structural responses, are termed as uncertain sensors henceforth. A sensor error function, one for each uncertain sensor, is formulated to detect the corresponding uncertain sensor failure in real-time, using the measurements from reference sensors and the uncertain sensor being monitored. The sensor error function is derived using the indirect and direct approaches. In the indirect approach, the error function is obtained from the state space model in combination with the inverse model and interaction matrix formulation. The input term is eliminated from the error function by applying inverse model and the interaction matrix is applied to eliminate the state and all uncertain sensor measurement terms excepted the examined uncertain sensor from the error function. In the direct approach, the coe±cients of the error function can be directly calculated from the healthy measurement data from the examined uncertain sensor and all reference sensors without having to know the state-space model of the system. Thus the need to know the state-space model of the plant can be bypassed. The sensor failure detection formulations are investigated numerically using a four degree-of-freedom spring-mass-damper system and experimentally using a 4m long NASA 8-bay truss structure. It is shown by means of numerical and experimental results that the developed sensor failure formulations correctly detect the instants of sensor failure and can be implemented in real structural systems for sensor failure detection.