In this study, we propose a knowledge-based approach for detection and isolation of sensor faults in fault tolerant control (FTC) of the three-tank system. Farthest first traversal algorithm (FFTA) of data mining is used first-time for the classification of faults in an FTC system. The sliding window is used to detect signal changes, which contain possible transients due to faults. The variance-changing ratio is calculated to extract the features of the sensor signal in each window. Then, FFTA is utilized for the isolation of sensor faults. In order to demonstrate the efficiency of the proposed method, seven types of artificial faults were applied to closed-loop fault tolerant control system in certain periods. All faults were detected and isolated immediately after they occurred. Moreover, fault isolation was achieved when multiple faults occurred simultaneously. (C) 2012 Elsevier Ltd. All rights reserved.