In order to compensate the destruction of hysteresis on trajectory tracking control accuracy of pneumatic muscle joint, a joint model is established. Furthermore, a hysteresis force equation is derived accordingly and the hysteresis characteristics are tested and analyzed. A model free adaptive CMAC (cerebellar model articulation controller) hysteresis compensation control algorithm is presented, which adopts the double structure of inflation and deflation. Gradient descent is employed to adjust network weights of the inflation and deflation processes. The fourth order Fourier fitting function is used to denoise network weights. Based on the Gaussian function and the neighborhood error, an error credibility evaluation function is designed to adjust the learning rate and suppress the impact of burst interference on the neural network. Then the triangular wave trajectory tracking control is applied to training the neural network. Finally, the trained neural network is applied to sine wave trajectory tracking control with sudden disturbance. Experimental results show that the control algorithm is self-adaptive to the hysteresis variation in nonlinear curves tracking control. Using this algorithm, not only the sudden disturbance is restrained effectually, but also the control accuracy is improved.