Knowledge-based approaches to software development promise to result in important break-through both regarding our ability to solve complex problems and in improved software productivity in general. A key technique here is to separate domain knowledge from control information needed for the procedural execution of a program. However, general-purpose inference mechanisms entail certain disadvantages with respect to e.g. efficiency, focusing in problem-solving, and transparency in knowledge representation. In this licentiate thesis we propose an approach, where domain-dependent control is introduced in the form of prototypes, based on typical cases from the application domain. It is shown how this scheme results in a more effective problem-solving behaviour as compared with a traditional approach, relying entirely on rules for domain as well as control information. Further we demonstrate how the knowledge base can be easily reused for independent purposes, such as consultative problem solving and teaching respectively. Our claims are supported by implementations, both in a conventional knowledge system environment with the help of the EMYCIN system, and in a system supporting reasoned control of reasoning, namely the economic advice giving in a bank environment, in particular advice on procedures for transfers of real estates.