For dealing with the problem that the estimate result of EKF is affected severely by the covariance matrices of noises, a new method using adaptive kalman filter(AEKF) to estimate simultaneously load torque and speed of motor is presented. while the speed and load torque are estimated，the input noise and noise introduced by modeling error are merged into an equivalent state noise in the augmented model of motor；Based on the transformed model, the covariance matrices of state noises are estimated using state predict residuals and the covariance matrices of observation noises are estimated using the measure residuals, then the adaptive diversification of the covariance matrices of noises are implement. The experimental results show that the estimate results are not affected by the given initial value of the covariance matrices of noises, and have high accuracy.