AbstractSome conceptual problems which are related to the way data (i.e. empirical information) are represented in model building are discussed. Elements of a framework for modelling are proposed in which a constructive relationship between the data and the modelling relation itself is incorporated. The main novelty of this approach is the use of the notion of information set, which is defined as the set of observed values of those observables which are actually available as data sources for a given model, as contrasted to a complete set of observations where questions of restricted availability cannot be raised. This permits the application of complexity theory in a new way. Some examples are given where, as a consequence, essential differences between descriptions, otherwise considered equivalent, will be revealed. One application, showing the existence of fundamentally inequivalent forms of dynamics which belong to different information sets, is elaborated in more detail.la science occidentale est fondée sur la doctrine monastique d'un univers ordonné, créé par un Dieu qui reste hors de la nature et la gouverne par des lois accessibles á la raison humaine.