The authors consider that in complex agrarian situations, decisions on land management changes should be based on a common understanding of the interactions between ecological/bio-physical and socio/economic dynamics that are at work. As agricultural and environmental issues are more and more inter-linked, the increasing multiplicity of stakeholders, with differing and often conflicting land use representations and strategies, underlines the need for innovative methods and tools to support their coordination, mediation and negotiation processes aiming at an improved, more decentralized and integrated natural resources management (INRM). During the past ten years, very significant advances in the simulation of societies in interaction with their environment have been achieved. More and more powerful and user-friendly computer modelling tools facilitate the understanding and simulation of such complex interactions. The main objective of this research is to study the use of MAS models, associated with role games, for knowledge integration in collective learning processes focusing on key local INRM issues. How can simulations be involved in such a processes, i.e., how can they help actors to govern the land? We are seeking to develop a companionable modelling use of multi-agent systems. For the past two years, the authors have used this methodology to address specific key INRM issues in northern Thailand and northern Vietnam. An example from Senegal is also presented and discussed.