In this work, speed sensorless state estimators are compared for induction machine drives. The studied estimators are based on two widely used state-space models. The first one has five state variables and assumes slowly varying rotor speed. In contrast, the second model is augmented by the equation of motion and the load torque is defined as an additional state variable. Due to the nonlinearities, extended and unscented Kalman filters are applied in the case of both models. To compare the parameter sensitivities and the low speed operation of the four state estimators, simulations and experiments are carried out. In addition, the estimators are also tested in speed sensorless closed-loop control structure.