Virtual reality offers a new frontier for human interaction with simulation models. A virtual environment, such as that created with a CAVE, imposes either real-time or quasi-real-time performance on the simulation model. Beyond that general requirement, what others can be identified for simulation programs that drive a virtual reality or virtual environment interface? Based on experience with the Virginia Tech CAVE augmented by a literature search, we propose three key requirements for successful CAVE-based simulations: (1) portability among CAVE-specific input/output devices, (2) effective and efficient interprocess communication, and (3) overcoming the limitations associated with input/output device interaction. Each requirement is described in some detail to both explain and justify its inclusion. Limitations and near- and intermediate-term research needs are identified.