This paper shows how a new approach in the use of AI techniques has been successfully used for the design of an effective ITS in the domain of diagnosis training. The originality of this approach was to take into account three complex problems simultaneously: teaching diagnosis methods to students, giving the means to the teachers of maintaining the system by themselves and providing a tool easy to insert in the context of university laboratories. The architecture of the system is based on a distinct use of two kinds of knowledge representation. All the knowledge liable to modifications is gathered within libraries under descriptive forms easily maintained by the educational staff. General diagnosis knowledge independent of hardware, circuits and even application fields, is described with basic production rules and control metarules. The development of the system was based on the precise analysis of the expert’s behaviour and of the user’s needs, with the aim of making extensive use of the descriptive forms in order to minimize the static knowledge embedded in the rules. The system can work on a microcomputer and is used in an engineering school.