Identifying tree species is of strategic importance for forest monitoring, analysis and management, and is crucial for sustainable forestry development. Therefore, we propose a method for tree species classification based on individual tree segmentation from airborne LiDAR, following by using the three-dimensional geometric features of the segmented tree crowns. First, TIN filter and tree points normalization are used to generate the Crown Height Model (CHM). Based on the geometric characteristics of the tree crowns, parallel-line shape fitting is used to fit the crown shapes. Specifically, three basic geometric shapes (triangle, rectangle and arc) are used to fit the crowns of different tree species. For the same crown shape or combinations of shapes, parametric classification is employed. The proposed method is tested using datasets from two different sites in the Tiger and Leopard National Park in Northeast China. The results show that the average accuracy of tree species classification is 90.9%, with the best accuracy reaching 95.9%, meeting the requirements of rapid forestry surveys.