Taking an improvement on the efficiency of aerial combat training evaluation as a target, and staring at the key nodes of influence on trend of combat outcomes, this paper proposes a method for determining aerial combat key nodes based on win rate prediction. Firstly, the long short-term memory networks are utilized for designing and training a win rate prediction model for real-time assessment of aerial combat situations. Secondly, the OTSU's method is utilized for determining the threshold for key nodes,and the time steps with absolute win rate difference between adjacent steps of being greater than the threshold are identified as key nodes. The experiments show that the win rates are accurate in prediction by the designed win rate prediction model, the adopted key node threshold selection method is reasonable, and the overall proposed method is feasible and has a good generality.