Abstract : The agronomists design many models whose usefulness, outside research, is questioned. This work, between agronomy and design ergonomics, proposes a methodology to design a tool DIAGVAR that could support cultivar assessment in link with its possible use by three groups of actors (seed breeders, people in charge of official assessment of the new cultivars and advisors). This design methodology aims at expanding jointly the agronomic model inscribed inside the tool and the activities of both designers and potential users. Our design methodology has combined an analysis of the activity of cultivar assessment and a stage of simulation where the potential users used a prototype of DIAGVAR on their own data. It has allowed us to identify many agronomic questions about the model of DIAGVAR. We have dealt with two of them, about new indicators of the N status of crops to make their measurement possible in cultivar trials and about the stability of statistical methods, used in the model of DIAGVAR, according to the various configurations of actors' databases. These questions have led us to strongly modify our agronomic model. From an ergonomic point of view, we have analysed the instrumental geneses caused by our design methodology. We have more globally analysed how our methodology modified the activity of the actors who took part to that work. We have finally proposed a frame to identify how our engineering project for the design of a tool allows us to produce scientific knowledge.