The present construction of global representations of food and farming is problematic. For example, how can we 'know' the world needs to double food production even though we cannot foresee a food crisis? How can we estimate investment opportunities while failing to quantify their impacts on smallholders? Global models constrain the manner in which we perceive the food regime while producing such representations. We need to identify the causal relations embedded inside models' equations and why they are arrayed in this fashion. This article combines actor-network theory and structuration theory to analyse a sample of 70 global models. It locates the modules and equations of these black boxes in the sociotechnical and political context of their production. Finally, a bibliometric analysis sketches the overall epistemic community that drove models into success or extinction. Dominant global models recycle equations, modules and databases to effectuate narrow worlds. They make smallholder farming invisible in spite of its prevalence around the world. They do not address food needs and construct pixellated representations of underutilized land. They systematically favour large-scale agricultural trade and investments in production and productivity. This reflects the structure of signification modellers adhere to as well as the structure of domination they are embedded in. Securing clients ensures the success of global models independently from their validation. The article demonstrates the manner in which modelling is a social practice embedded in power relations. Considering simultaneously the structure of domination formalized inside models and surrounding modelling is crucial. Future research should investigate how various actors resort to global models to champion their goals. It should question the policy recommendations drawn from such models and their relevance as decision support tools.