Abstract: A neural-network-based real-time IP position control scheme for permanent magnet synchronous motor PMSM) was proposed. According to the IP control scheme of servo system, a neural-network real-time IP position controller was implemented by the superior approximate performance of neural network. This controller has simple structure, reduces computation amount of weight adjustment and speeds up the computation. A hybrid neural network was applied as identificator of PMSM system, and an algorithm under multi-step predictive index function was used to train neural network on-line. The load torque was acquired from the identificator, and fed forward to compensate the load torque disturbance. The proposed strategy has high-performance dynamic characteristics and strong robustness.