This paper provides a light introduction to a simple graphics technique which can be used to represent random data on a graphics supercomputer. The representation, called a "noise-sphere," can be used to detect "bad" random-number generators with little training on the part of the observer. The system uses lighting and shading facilities of 3D extensions to the X-Windows or the PHIGS+ standard. To encourage reader involvement, computational recipes and suggestions for future experiments are included.