Abstract: This paper presents a new impostor-based approach to accelerate the rendering of very complex static scenes. The scene is partitioned into viewing regions, and a layered impostor representation is precalculated for each of them. An optimal placement of impostor layers guarantees that our representation is indistinguishable from the original geometry. Furthermore the algorithm exploits common graphics hardware both during preprocessing and rendering. Moreover the impostor representation is compressed using several strategies to cut down on storage space.