Abstract
After a brief presentation of what I believe to be the main features of the modelling démarche in science, I will focus on the basic following question: how can an abstract entity - a model - possibly represent an existing observable entity, which is phenomenally accessible to us, but which is not abstract? This is what Bas van Fraassen calls the loss of reality objection. Instead of proposing a pragmatic dissolution of this objection as van Fraassen does, I will argue that scientific representing necessarily involves a homomorphism between structures and thus that, in a strict technical sense of representation, phenomena are not represented by our models but only as having some specific properties belonging to the domains of our representing structures. Yet, our contact with phenomenal entities is secured by propositions that are true in a correspondence sense. The truth of these - ontic - propositions which state that phenomenal entities possess some precise properties is the soil on which the informative content of our successful representations rest.