In order to solve the contradiction between high guidance accuracy and fast real-time solving in traditional predictive guidance method, a neural network predictive guidance method is presented, based on the pattern of optimal guidance. The method predicts trajectory based on high believable simulation of kinematic aircraft model, and uses optimization theory to iterative solution of guide variable, so as to generate off-line sample data. By means of choosing multi-modal neural network, training neural network based on dispatching management, to complete the design of the neural network prediction guidance controller. CAV as an example to design, results show that: The method is less real time calculation, fast real-time solution and high guidance accuracy, of which the comprehensive performance is far better than the traditional predictive guidance method.