To address the problem that the existing trunk extraction methods are affected by sagging branches and leaves as well as noise, this paper proposes a hierarchical trunk segmentation method based on the normal statistical features of point clouds. The trunk is firstly nested to preserve the complete trunk while removing the sagging branches and leaves. Furthermore, the trunk is layered, and the point cloud is statistically filtered for each layer. Finally, the normal vector of each layer is calculated to calculate the correct segmentation height based on its statistical features. Experiments were conducted with two trees, Ginkgo biloba, Metasequoia and willow, and the coefficient of determination were 0.974, 0.934 and 0.922, respectively, and the root mean square errors were 0.070 m, 0.075 m and 0.132 m, respectively, which indicates that the method has high segmentation accuracy and can provide technical support for the accurate extraction of tree structural parameters.