A theoretical model specifying the underlying knowledge and procedures whereby human subjects can generate effective initial descriptions of scientific problems was formulated. The model is prescriptive since it does not necessarily try to simulate the behavior of actual experts nor assume that their performance is optimal. The model, elaborated in the domain of mechanics, specifies explicit procedures for redescribing problems in terms of a relevant knowledge base. To test the model, carefully controlled experiments were devised where human subjects were induced to act in accordance with alternative models and where their resulting performance was observed in detail. Such experiments, carried out with undergraduate physics students, showed that the proposed model is sufficient to generate excellent problem descriptions, that these markedly improve subsequent problem solutions, and that most components of the model are indeed necessary for good performance. Detailed data analysis also showed how the model predictably prevents the occurrence of many common errors. Such a validated model of effective problem description provides a useful basis for teaching students improved scientific problem-solving skills.