Aimed at the problems that the traditional evaluation methods are low in evaluation precision, complex in structure of the model, and is not available in the real-time dynamic threat assessment, an air combat threat assessment method based on least squares support vector machine (LSSVM) is proposed. Firstly, the threat index of air combat characteristic data is analyzed, and the expert evaluation is used to build the threat assessment sample base. Then, cross hybridization chaos quantum particle swarm optimization(CHCQPSO) algorithm is used to optimize the regularization parameter and kernel function parameter in LSSVM, and the results are compared with those of the classical PSO, BP neural networks and mesh models. Finally, the optimized LSSVM model is used to realize the real-time dynamic threat assessment of air combat targets. The simulation results show that the proposed method is high in accuracy and short in time required, thus providing a new idea for evaluating the threat of air combat targets.