In the permanent magnet synchronous motor (PMSM) AC servo control system, the phase current shows nonsinusoidal variation due to the switching of electronic devices. The direct and quadrature axis currents obtained by coordinate transformation contain abundant harmonic components, causing current loop oscillation and electromagnetic torque ripple. On the basis of analyzing the relation between the phase current and the higher harmonics of the direct and quadrature axis currents, as well as the effect of higher harmonics on the electromagnetic torque, a feedforward control strategy based on adaptive FIR filter was proposed for AC servo system. Furthermore, a variable step length factor algorithm was proposed, with which the step length factor was changed according to the error value, so as to improve the convergence speed of filter and reduce the overshoot. Compared with the traditional firstorder lowpass filter, this algorithm had no phase lag, and showed higher sensitivity and system stability. The improved control strategy was effective to reduce the direct and quadrature axis current harmonics, restrain the current loop oscillation and electromagnetic torque ripple, and improve the PMSM AC servo control system stability and dynamic response. The feasibility of this method was proved by simulation and experiment.