In this paper, a new model-based fault detection and diagnosis method for broken rotor bars in squirrel-cage induction motor is proposed. The proposed method relies on innovation sequence generated by the conventional extended Kalman filter. The innovations would follow a Gaussian distribution under normal operation; however a fault, i.e., broken rotor bar, would change this underlying distribution. It has been shown that this change in the distribution is indicative of a fault. The proposed method uses readily available current measurements and no additional sensors are required. Further, the proposed method is robust to unbalanced supply voltage and load changes. Computer simulations are carried out for 4-hp squirrel-cage induction motor using MATLAB software. The results demonstrate the advantage of the proposed technique as it provides accurate estimates for broken rotor bar fault detection. © 2016 IEEE.