An adaptive speed PID controller based on RBFNN is investigated for PMSM servo system in this paper. RBF is used to identify the plant first. Then the Jacobian information of controlled plant obtained by RBFNN identification is used to adjust the gain of speed PID controller on-time. A robust parameters self-tuning learning algorithm is adopted to train the ANN for fast convergence. Simulation result shows that the proposed controller has the adaptability, strong robustness and satisfactory control performance in PMSM servo system.