This study presents a parameter tuning approach for Extended Kalman Filter (EKF) based observers for the sensorless control of Induction Motor (IM) drives. After an analysis performed on the effect of covariance matrix elements of EKF, the study demonstrates the improved performance of the EKF based estimation (performed for stator currents, rotor flux, rotor speed, stator resistance and load torque), via the developed online parameter tuning approach for different speed and load references. Firstly, it has been demonstrated experimentally that covariance matrices used in EKF algorithm vary with the operation conditions. It has specifically been demonstrated that, among the elements of model covariance matrix, the ones corresponding to the rotor flux components are the most effective in correcting the estimations of the related EKF algorithm. To address this issue, an online fuzzy approach is developed based on different load and speed references, of which the inputs are the estimated speed and estimated load torque, and the output consists of the elements of the model covariance matrix related to the rotor flux. The performance of the proposed Fuzzy EKF has been experimentally tested and the results have demonstrated that the proposed scheme can eliminate biases and yields higher estimation accuracy when compared with the standard EKF where the tuning parameters are fixed to constant values.