Considering the three-dimensional(3D) terminal guidance issue, based on the neural dynamic surface control theory, a novel guidance law is designed, and its stability has been proved by Lyapunov stability criterion. The radial basis function(RBF) neural networks are used to estimate the target motion information in real-time and reduce the coupling interference in the 3D motion model. Thus the guidance law can be self-adaptive to the variation of parameters, which improved its robustness. Simulation results shows that the missile which uses this guidance law can obtain smaller miss distance in the case of target maneuvering complicated.