Based on analyzing the factors that affect air-targets threat assessment, an air-targets threat assessment model is established based on the target threat value, target threat capability, and target threat level. To solve the problem that the dolphin swarm algorithm is easily trapped into local optimal solution and appears premature convergence, the chaotic dolphin swarm algorithm(CDSA) is proposed on the basis of the basic dolphin swarm algorithm, by making use of chaos to improve the initialization, dynamic clustering and premature optimization. The method employing the chaotic dolphin swarm algorithm to seek the global excellent result to simultaneously optimize the initial weights and thresholds of the grey neural network(GNNM) is presented. And on the basis of it, an air-targets threat assessment model is established. Compared with the GNNM and DSA-GNNM, the simulation results show that the CDSA-GNNM not only improves the global optimization performance, but also obtains robust result with good quality. Through simulation and analysis of experimental data, the effectiveness of the proposed algorithm in the application of air-targets threat assessment is verified.