Abstract: As a kind of optimal control method, Model predictive control（MPC） attracts much attention in the area of AC motor drives recently. For two-level inverter-fed induction motor（IM） drives, conventional MPC needs to predict both stator flux and stator current for seven times to obtain the best voltage vector minimizing the cost function, which leads to high computational burden and is unfavorable for real-time implementation. Based on the analytical study of the relationship between stator flux and stator current, this paper proposed an improved MPC, which only needs to predict the stator flux and avoids the complicated stator current prediction. Hence, the control complexity and computational burden are significantly reduced. Furthermore, to improve the system reliability and reduce the hardware cost, speed adaptive full order observer was introduced to achieve sensorless operation. The influence of control delay was compensated and the strategy of pre-excitation was implemented to avoid large starting current while obtaining high starting torque. Finally, simulation and experimental studies were carried out on a two-level inverter-fed IM drive. The results prove that the proposed sensorless MPC can achieve excellent dynamic and steady state performance over a wide speed range.