Architectural scenes usually have some peculiar properties, like parallel surfaces, symmetry, instantiation, easy clustering, etc. These properties can be used by visualization algorithms in order to improve the solution of the hidden surface removal problem.
This paper presents some practical implementation policies for the binary space partition algorithm proposed by Fuchs. They take advantage of the characteristics found in architectural scenes. These suggestions help to minimize both the time needed to build the tree and the memory space used. Five optimization techniques are presented and applied to fourteen architectural scenes.