In many virtual reality applications as well as general computer graphics we need to consider large numbers of objects to render one image. In many cases rendering can be preceded by a culling phase that employs simple mechanisms to reject most of the objects. As a result, only a very small portion of the model has to go through the time-consuming process of hidden object removal. We report on such a culling mechanism that is based on regular space subdivision into cells followed by cell classification into interior, exterior, and wall cells. A special cell-to-cell visibility algorithm is then activated between every two nonexterior cells. Only the objects in the potentially visible set of cells are actually submitted to the hidden object removal algorithm. We report on the implementation of the algorithm and its performance for walkthrough of various environments.