Many important computer applications require that users be able to use them effectively with little or no formal training. Current examples include bank teller machines and airport information kiosks. Today successful systems of this kind can only be developed by iteration using costly empirical testing. This article aims to provide a theoretical foundation for the design of such systems, a model of learning by exploration called CE+. The theory incorporates assumptions from (a) the GOMS model and cognitive complexity theory (CCT) on the representation of procedural knowledge as productions, (b) the EXPL model on learning from examples, and (c) research on problem-solving processes for simple puzzlelike problems. Design guidelines for systems that can be learned by exploration, "design for successful guessing," are derived from the theory. These principles are compared with those developed by Norman (1988).