At the crossroads of the ethnography of scholarly practices and digital humanities, this study proposes to consider the exploratory aspect of the scientific activity as an intellectual path through a corpus made accessible by an online platform. My case study focuses on the reading paths of Gallica users by combining a qualitative approach—semi-directive interviews and research situations—with a quantitative approach—the analysis of one year of server logs of the platform by extracting paths, modelling them by Markov chains, and identifying typical paths by topological analysis of the data. This method shows the great complementarity of the qualitative approach and the quantitative approach: by confronting the material conditions of the possibilities of navigation with its real material practices of appropriation, in specific social and cultural contexts, this project strikes a middle ground between technological determinism and constructivism. This exploratory study has led to promising results on the navigation of researchers within the Dewey classification, on the role of “pivotal literature” in moving from one discipline to another, and on a first sketch of a variety of “regimes of navigation” grouped according to their topological characteristics. These initial results have made it possible to document in detail the intricacy of material practices and virtual interfaces, as well as the navigation strategies developed by the researchers, which testify to their desire to find “original branches”. I have therefore been able to highlight the non-linearity and indeterminacy of scholarly practices and the fundamental nature of navigation as an orientation practice. But such an understanding of digital libraries as spaces, and their navigation, must not be without a strong epistemological and methodological critique. Recognising, on the one hand, the reification—or the effet de reel—produced by visual representation, digital libraries cannot be considered as uncharted territories just waiting to be mapped. Distrusting, on the other hand, the renewed fantasies of a “cyberspace” that could be constructed from scratch, I would like to argue that it is possible to find a middle way between the referential illusion and the demiurgic fantasy and, in this dehiscence, to deploy a practical use of cartography as an object that performs a space of knowledge, but that performs it under constraint and in a critical and reflexive way.