Designing or improving farming systems requires understanding their dynamics so as to predict their behaviour in response to management. Simulation tools can potentially support the process by which farmers and scientists might obtain such an encompassing understanding. The usability of these tools is, however, partially inhibited by the inherent complexity of the interactions at work in farm-scale models. Whereas such models are generally used in isolation, here we present an approach in which a field-scale diagnosis method complements a farm-scale simulation model. This diagnosis method lends itself easily to an intelligible presentation of field-specific knowledge that can be fed to the simulation tool for more encompassing considerations. Our approach is used to support the design of novel management strategies in grassland-based beef systems and proved to be effective when applied to two farms in the French Pyrenees. Thanks to the integrative representation of the various processes, including the management ones, simulation contributed to deeper learning of both scientists and farmers about room for manoeuvre for increasing self-sufficiency for forage. The diagnosis phase enhanced the learning process by providing a simpler framework in which elementary problems at field scale could be considered separately before being examined concurrently at farm scale in the simulation phase.