There are many different classes of decision made during the development of a model. If models are to be shown to be appropriate for the specified task, the information and assumptions upon which these decisions are based must be made explicit. This paper describes AIM, an automated modeller that makes all its decisions with explicit knowl-edge. AIM separates different types of modelling knowl-edge in its knowledge base to prevent the embedding of as-sumptions in the structure of the modelling knowledge. It also uses a novel modeling algorithm that does not rely on either external sources of information or the structure of the knowledge base to generate models. We show that AIM generates parsimonious models and discuss some of the constraints on this approach.