A prescriptive model of problem-solving performance is described. The aim of the model is to specify explicitly how problems should be described initially in such a form as to facilitate their subsequent solution. According to the model, the process of initial problem description can be decomposed into two successive stages. The first uses general domain-independent knowledge to generate a "basic description" of a problem, the purpose of which is to identify explicitly the information specified and wanted in the problem, to introduce useful symbols, and to express the relevant information in various convenient symbolic representations. This basic description is then used to generate a "theoretical description" of the problem, that is, a deliberate redescription of the problem in terms of special concepts provided by the domain-specific knowledge base, involving the particular entities of interest in the problem, describing these entities in terms of concepts specifically provided by the knowledge base, and exploiting the known properties of these descriptive concepts. Focusing on mechanics (physics), the knowledge needed to generate theoretical problem descriptions is discussed, followed by experimental methods for testing the prescriptive model and selected results obtained from such experiments. (Author/JN)