Cerebella Model Articulation Controller (CMAC) is considered as local association and generalization neural network. Parametric CMAC (P-CMAC) is a modification to the original CMAC. The introduction of continuous activation function applied in the input space can overcome the binary behavior of the original CMAC. Takagi-Sugeno(TS) type fuzzy inference is embedded in the internal mapping to improve the approximation accuracy. Hybrid control scheme, which combines P-CMAC neural network and traditional PID controller, is proposed in the paper. The output of P-CMAC network dominates the overall control signal applied to the plant, while the traditional PID controller serves as compensator for reducing tracking error. The application of hybrid control scheme to filament tension control is illustrated. The experimental results have shown the effectiveness and accuracy improvement of the control scheme.