The maneuver mode recognition problem of reentry gliding vehicle (RGV) is the key to the interceptors in achieving its trajectory prediction. In view of this issue, this paper proposes a set of feature parameters fitted to the maneuver characteristics of vehicle trajectories. Based on the constructed RGV maneuvering mode trajectory library, an LSTM deep learning neural network is built, training the extracted new feature parameters. Compared with the traditional modes recognition method and other typical feature parameters in network training, the results show that the set of the proposed feature parameters is fast at convergence speed, high in recognition accuracy, and good in robustness in LSTM maneuver mode recognition network training.