Laser point cloud data contain detailed and high precision information, which is widely applied to forest evolution and plant reconstruction. In order to improve the accuracy and photorealistic in tree 3D reconstruction, a 3D reconstruction method based on point cloud was proposed. Firstly, the Kinect 2.0 was used for acquiring double face point cloud data of tree, and there were four plastic balls which were artificial markers near the tree. In the process of registration, artificial marker method and iterative closest point (ICP) algorithm were combined to merge point cloud data together accurately. Then, the complete point cloud data of tree can be obtained. Secondly, the improved space colonization algorithm was used with growth angle constraint to generate the 3D skeleton of tree, and then thickness of tree was estimated according to pipeline model, and the trunk was generated by generalized cylinder. Finally, leaves which were modeled based on point cloud data were added to branches according to phyllotaxis. Reconstruction experiment was performed by using magnolia, maple and limit tree data, which was provided by Visual Computing Research Center of Shenzhen Institutes of Advanced Technology (VCC). The reconstruction results showed that the proposed method can simulate 3D morphological structure of tree realistically and it showed the topological structure of trees well. Reconstructed tree models were plausible to the real-world trees and reconstruction error was less than 6.5%. Meanwhile，the proposed algorithm could be applied to the researches of three-dimensional modeling of virtual trees, virtual pruning experiment and tree topological structure analysis.