As the demand for high levels of interaction in computer systems increases, so too does the need for real-time, interactive animation. Detecting collisions between geometrically modelled objects remains a major bottleneck in areas such as Virtual Reality (VR). In order to maintain a constant frame-rate, a trade-off between speed and accuracy is necessary. This is possible if, at each frame, potential collisions are graded by their importance to the viewer’s perception. An appropriate Level Of Detail (LOD) at which to test each object may then be chosen, based on the importance of the collision in which it is involved. We adopt some ideas from an emerging area of research, Interactive Vision, and propose a scheme which uses an eye-tracking device to locate the position of the user’s gaze. This, along with other perceptual criteria, may be used to choose an appropriate LOD for each colliding object at each frame, allowing the application to degrade detection accuracy where it is least likely to affect the user’s perception of the collision.