In the domain of natural resource management or land planning, companion modelling (Bousquet et al. 1999) has built up as a recognized approach to use models for interacting with local communities and / or decision makers about the structure, dynamics and evolution of their territory. The counterpart of this approach that builds up ad-hoc models strongly tightened to local stakeholders stakes and perspectives is that generalisation or comparative works are difficult. However, even if all case studies and situations are unique, generalisation and comparison can be very useful in 1) raising the level of knowledge building by allowing to question one case studies with hypotheses from another one or expanding scenarios space 2) speeding and easing model building by providing design patterns. The idea of our work is that a generic model can be used as a basis for comparing case studies and situations where a similar question is raised. We develop and test a methodology for incorporating a generic model development in companion modelling cycles over 5 case studies where the question raised is: what are the interdependencies related to water between the rural and urban components of a territory? We consider both interdependencies that are linked to the water resource, extended to the services it can procure, naturally or through infrastructures, and to the functions it may have in an ecosystem, and interdependencies that are linked to institutional devices. We consider that these interdependencies are constrained by topological and morphological traits of the geographical, cognitive and social spaces where concerned actors and resources live. As a consequence we build up a generic model where root entities are : Actor, Resource, Tool, Interaction and Space. The design and refinement of this generic model is initially done by getting simple archetypal models or thesaurus from domain experts. The development of the ad-hoc case studies models will then be closely followed as these case studies models will be used to populate our generic model. This exercise will question the generic model that will question in return the case studies specific models. We expect that after a few loops we will be able to have specific models and the generic one converging. In this way we will have managed the different case studies models to share a common vocabulary and a common meta-structure. This will allow to ease comparison by observing where the different models belong in the common meta-structure.