In archaeological investigation such as analyzing posthole remains and other fragmentary evidence to infer and reconstruct the superstructure, the process of inference used to reconstruct the past varies among researchers. Therefore, it is essential to delve deeper into discussions, including the validity of these inferences. At that time, by creating 3D models of the superstructures and engaging in thorough discussions, consensus can be reached, resulting in more accurate reconstruction proposals. In this study, based on nearly regular building polygons, we focus on ancient buildings such as multi-tiered pagodas and ancient gates, particularly the substructures beneath the roofs. Some parts of the substructure take complicated shape, which cannot be formed through simple CSG technique which employs Boolean operations on primitive solids (such as box, prism). For complicated parts creation, we use “descriptive geometry” which represents 3D objects in 2D view by theoretically using planar geometric projections. We automate the generation of these structures by automatically unfolding front view (outer contour line of the structure) onto the xy-plane, extruding them, then unfolding the side view, and through Boolean operations between two “extrusions”, integrating these structures into the buildings, resulting in automatic generation of 3D ancient building models.