Due to the complexity of real environmental acoustics, Virtual Reality applications have to use a simplified model, in order to achieve realtime processing of the sound information. This simplification in rendering the sound field can lead to a less realistic experience for the user as subtle information is lost or misrepresented. This paper discusses a two stage technique for simulating room acoustics. In the first stage an accurate model of the room's impulse response is generated along with a graphical model on the environment. This model is calculated off-line, to avoid the time constraints that affect realtime processing. In the second stage the user explores the environment in realtime. Graphical and auditory information appropriate to the position and orientation of the user is generated from the results calculated in the first stage. The aural results are positioned using ambisonics to produce a spatialised 3D sound output from a set of loudspeakers. By removing the rendering of the impulse response from the realtime stage we can consider a large number of reflections and model the late reflections accurately. The technique has been applied to the modelling of possible improvements to a real multimedia presentation room, which presently exhibits poor acoustics.