We propose a novel hybrid representation scheme for indoor environments using meshes and point clouds. Points are suitable for rendering detailed objects, while meshes can represent planar shapes more compactly. Based on these properties, we use points or meshes adaptively according to local characteristics of an input scene. More specifically, we first determine building structures, which can be represented by meshes, in indoor environments. Then, we create meshes corresponding to the building structures, while representing the remaining parts with points. Experimental results demonstrate that the proposed hybrid scheme renders indoor environments more efficiently than the point-based representation or the mesh-based representation.