This paper describes a novel from-region visibility algorithm, the unique properties of which allow conducting remote walkthroughs in very large virtual environments, without preprocessing and storing prohibitive amounts of visibility information. The algorithm retains its speed and accuracy even when applied to large viewcells. This allows computing from-region visibility on-line, thus eliminating the need for visibility preprocessing. The algorithm utilizes a geometric transform, representing visibility in a two-dimensional space, the dual ray space. Standard rendering hardware is then used for rapidly performing visibility computation. The algorithm is robust and easy to implement, and can trade off between accuracy and speed. We report results from extensive experiments that were conducted on a virtual environment that accurately depicts 160 square kilometers of the city of London.