Over the last few years, object-oriented techniques have gained an increasing recognition both in software engineering and in AI. Object-oriented systems present undisputable advantages and provide features that make them really suitable to represent knowledge. However, our practical experience in using these techniques for knowledge representation led us to discover that objectoriented systems also present serious drawbacks, essentially due to their lack of expressive power. These drawbacks can really be felt when modelling domains characterized by a wide variety of knowledge. This paper introduces the notion of Representation Cluster, which allows to provide any object-oriented system with customizable knowledge representation formalisms. These formalisms enable to express and handle diverse kinds of knowledge using only their natural highlevel properties. Such an approach speeds up knowledge bases development, makes them clearer, natural, and avoids a great deal of code writing. It...