This thesis is a study of scientific practice in computational physics. It is based on an 18 month period of ethnographic research at the Imperial College Applied Modelling and Computation Group. Using a theoretical framework drawn from practice theory, science studies, and historical epistemology, I study how simulations are developed and used in science. Emphasising modelling as a process, I explore how software provides a distinctive kind of material for doing science on computers and how images and writings of various kinds are folded into the research process. Through concrete examples the thesis charts how projects are devised and evolve and how they draw together materials and technologies into semi-stable configurations that crystallise around the objects of their concern, what Hans-Jorg Rheinberger dubbed “epistemic things”. The main pivot of the research, however, is the connection of practice-theoretical science studies with the philosophy of Gaston Bachelard, whose concept of “phenomenotechnique” facilitates a rationalist reading of scientific practice. Rather than treating reason as a singular logic or method, or as a faculty of the mind, Bachelard points us towards processes of change within actual scientific research, a dynamic reason immanent to processes of skilled engagement. Combining this study of reason with the more recent attention to things within research from materialist and semiotic traditions, I also revive a new sense for the term “representation”, tracing the multiple relationships and shifting identities and differences that are involved in representing. I thus develop a theory of simulation that implies a non-representationalist concept of representing and a non-teleological concept of reason.