This paper does not focus on models proper, but on the notion of modelisation as a creative process and a solution to one of the key issues of XIXth century’s intellectual life : how can one grasp both the one and the many at the same time ? The answer lies in two specific areas within which naturalism can be considered “abstract” : the theory of characters, as redefined by Zola through a statistical approach of “types” (Quetelet) ; the graphic sketches that served as the bases of Rougon-Macquart’s genesis, and make scientific models appear as fiction-shifters. At the intersection of these areas, the formal play unveils itself as a so-said scientific justification for reactivating popular knowledges despised by official science. Zola reveals that models are fictional, and that paradoxical though it may be, they can be accurate representations precisely because they are approximate and non-realistic. These we will define as non-euclidian fictions.