To satisfy varying demands of spatial information in metropolitan area, three-dimensional models of building interiors are oftentimes desired as an indiscernible component of geographic information system. In this paper, a semi-automatic approach consisting of segmentation and outline tracing processes is introduced to construct indoor three-dimensional models from the terrestrial LiDAR. In the segmentation process, the random sampling consensus algorithm and virtual grid are used to group point clouds that belong to identical planar planes as well as to remove outliers from point clouds. In the outline tracing process, a data conversion and a skeleton algorithms based upon a virtual grid are used to extract lines from those segmented point clouds. However, despite of an improvement of productivity, the proposed approach requires an optimization process to adjust parameters such as a threshold of the random sampling consensus and sizes of the virtual grids for a filtering and a line extraction. It is required to determine those parameters in a heuristic way, depending on the characteristics of indoor environments. Furthermore, the proposed approach needs to be devised to model curvilinear and rounded shape of the indoor structures.