A neural control method based on fusion of two models was proposed for a quadruped robot with multiple degrees of freedom, combining central pattern generator (CPG) and model predictive control (MPC). A behavioral movement neural control architecture for a legged robot was constructed based on model predictive theory by simulating biological neural mechanisms. This architecture can process the external environment information, adaptively adjust the position of the body and legs, and realize position tracking, omnidirectional movement and a variety of atypical gaits of quadruped robot. The experimental results show that the quadruped robot based on the MPC-CPG architecture can quickly respond and eliminate the position error and angle error, the position error in trajectory tracking is always kept at ?0.1~0.1 m, and the attitude angle error is kept at ?0.05~0.05 rad. The quadruped robot not only has high trajectory tracking accuracy, but also exhibits behavioral diversity with the MPC-CPG controller, which verifies the effectiveness of the proposed MPC-CPG controller.