Aiming at the shortcomings of the traditional clustering algorithm on the clustering effect of manifold data, and the low real-time performance, the improved CBSCAN (cell-based density spatial clustering of applications with noise) was proposed to solve air combat target grouping issue. By analyzing the air combat situation parameters, the general model of air combat target grouping was established and the target grouping was transformed into clustering problem. Then, the target grouping model based on improved CBSCAN was established by improving the clustering method of CBSCAN algorithm. Through simulation experiments, the clustering accuracy and real-time performance of K-means, expectation maximum algorithm, density peak algorithm, density-based spatial clustering of applications with noise algorithm, CBSCAN algorithm and improved CBSCAN algorithm in 30 combat situations were compared and analyzed. The results show that the improved CBSCAN algorithm can correctly group multi-target formations under the condition of unknown number of formations and target manifold distribution, and the real-time performance was improved by about 30% compared with the original algorithm, which shows the practical application value of the proposed method.