I treat three related subjects: virtual-worlds research—the construction of real-time 3-D illusions by computer graphics; some observations about interfaces to virtual worlds; and the coming application of virtual-worlds techniques to the enhancement of scientific computing.We need to design generalized interfaces for visualizing, exploring, and steering scientific computations. Our interfaces must be direct-manipulation, not command-string; interactive, not batch; 3-D, not 2-D; multisensory, not just visual.We need generalized research results for 3-D interactive interfaces. More is known than gets reported, because of a reluctance to share “unproven” results. I propose a shells-of-certainty model for such knowledge.