A speed and flux estimation method of induction motors using fuzzy extended Kalman filter (FEKF) is proposed in this paper, which is used to make lower impact of time varied statistic of measurement noise. It reaches a better speed estimation accuracy of induction motors than the extended Kalman filter (EKF). The proposed algorithm modifies the measurement noise covariance of extended Kalman filter recursively by monitoring if the ratio between filter’s innovation and actual innovation is near 1, and chooses a fuzzy factor to make its noise model close to real noise model adaptively. The speed estimated error and the flux fluctuation of FEKF under gross external disturbance and unknown measurement noises are compared with EKF. Simulation and experimental results show that FEKF provides better performance and faster convergence than EKF under gross external error and unknown measurement noises.