Abstract: This paper presents a novel method for computing visibility in 2.5D environments based on a novel theoretical result: the visibility from a region can be conservatively estimated by computing the visibility from a point using appropriately "shrunk" occluders and occludees. We show how approximate, yet conservative, shrunk objects can be efficiently computed in an urban environment. The technique provides a tighter potentially visible set (PVS) compared to the original method in which only occluders are shrunk. Finally, theoretical implications of the shrinking theorem are discussed, opening new research directions.