Constructing a knowledge based system requires the identification of the conceptual primitives and problem-solving strategy that will permit capturing the expertise in an adapted way. This paper presents an approach to support the elaboration of such a conceptual model by data-abstraction. Our work is motivated by the objective of constructing models that correspond to how experts solve problems, which is required for some applications, as, for example, presenting problem-solving in an educational system. We argue that such an objective cannot be tackled by refining a predefined model. We propose to use a prototype reifying the model as a means for its study. We present methodological guides, and an operationalisation language that allows the putting into practice of the approach we propose. Examples from different applications illustrate the presentation. (http://www-lium.univ-lemans.fr/~tchou/IJIS-97.pdf)