Abstract: Tree-like structures, recognized for their mechanical efficiency and aesthetic appeal, have extensive applications in long-span steel structure buildings. The mechanical performance of such structures greatly relies on the rationality of their forms. Hraditional form-finding methods encounter limitations including low efficiency, convergence difficulties, and poor adaptability to complex loading conditions. To address these issues, this study proposes a data-driven intelligent form-finding method for tree-like structures by employing machine learning techniques to achieve rapid and efficient configuration design. The basic principles of three representative machine learning methods—BP neural networks, RBF neural networks, and random forests—are introduced. A data-driven framework for intelligent form-finding of tree-like structures is developed. Four typical cases, including secondary branching, tertiary branching and, three-dimensional planar roof, and cylindrical latticed shell tree-like structures, are analyzed and compared with the inverse hanging recursive method to verify the effectiveness of the approach proposed. The research results demonstrate that the data-driven intelligent form-finding method can significantly reduce member-end bending moments, is applicable to two- and three- dimensional, and to complex spatial cylindrical latticed shell structures, and performs well under various loading conditions. This study provides a new perspective and methodology for the intelligent design of tree-like structures.