Multi-sensor monitoring devices that use skin surface or implanted sensors are susceptible to changes in temperature, sweat, and movement, such that the measured data cannot be used. This paper presents an automatic approach to detect such erroneous sensors. It is based on the assumption that valid measurements are related by a reconstruction model, while measurements from erroneous sensors are unrelated. The method estimates the data at each sensor based on the measurements from all other sensors, and compares it to the measurements. The sensor-data match is tested using ANOVA to detect the presence of an erroneous sensor. The method was tested on simulated and experimental data of Electrical Impedance Tomography (EIT) and showed consistent identification of erroneous electrodes.