Abstract: The difficulty of 3D modeling for standing trees is to simplify complex and scattered mass measurement data. In regard to individual standing tree sample, the geometric shape of trunk is extremely irregular and unstructured, branches and leaves grow optional and in dispersion, and the trunk measured point cloud data(PCD)obtained by the terrestrial laser scanning(TLS)LiDAR is extraordinary numerous and jumbled. In this paper, we propose a new skeleton-based 3D modeling method for tree branches based on terrestrial laser scanned PCD. Firstly, leaves and branches of the original PCD are separated using semi-supervised SVM classifier. Then, the PCD of branches are segmented according to Dijkstra distance, and the skeleton of each connected part is extracted. After calculating the weights of distance and angle by linear programming, the skeletons are connected according to the weighted matching degree. The complete skeletons of the whole tree are done, and the models are reconstructed by cylinder fitting. Experiments were carried out on Sakura and Michelia maudiae with the models reconstructed and the effectiveness of the algorithm analyzed. The results showed that the method used in this study is better than other previous methods in running time and occupying memory.