Abstract: This paper studies the dynamic structure of a type of compound fuzzy neural network, which can effectively use the process knowledge. The network has the ability of dynamic mapping by adding recurrent nodes in the second layer of the function network of the compound fuzzy neural network, so it would realize the good response to the dynamic system. Simulation has been made with a dynamic nonlinear model, and the result shows that the compound dynamic fuzzy neural network is better than the static network in convergence rate and prediction precision, and has a smaller network size.