In the 1970s Mali's Institut d'Econonzie Rurale (1ER), like many research institutes, created a Fanning Systems Research (FSR) department with the aim of making research more responsive to farmers' needs. While this department itself had much success its approach failed to influence other departments and to /make the entire Institute more client-oriented. Building upon the lessons of successful agricultural research in western countries 1ER has now embarked upon a bold new initiative to establish partnerships with farmers' organisations. This paper describes the process of establishing such partnerships, highlighting the role of service NGOs and the solution identified to the difficult issue of representation. For research purposes it is important for different agroecological and socioeconomic conditions to be represented. While the vehicle through which this is done may be a farmers' organisation, it is the underlying conditions rather than the actual organisation which must be represented. The paper also highlights the changing political climate in Mali, in particular the move towards decentralization, and the institutional adjustments which are making the new partnerships feasible, namely: (i) changes in the governance of 1ER to make the institution accountable for its results; (ii) changes in scientists' directives so that they too are held accountable for results; (iii) a move to more formal programme planning and review, enabling farmers' inputs to be incorporated. In addition, a pool of money has been made available to the users of research to enable them to contract research on topics which are of direct interest to them. It is hoped that this will *increase the farmers' influence and thereby strengthen their partnership with researchers. Although the experiment is in its early stages, two important lessons can already be drawn. The first is that fanners must be involved at all levels of the decisionmaking process. If they are represented only at the national level then the partnership will lack a practical base and will risk becoming a partnership in name only. If farmers are represented only at a local level, they will be excluded from crucial decision-making about resource allocation and long-term research priorities. The second lesson is that farmer representatives must participate at all stages of technology generation i f joint programme and review meetings are to be effective. They should therefore be included in: constraint diagnosis; identification of possible solutions; screening of technologies on station and test and evaluation of technologies on-farm. Hence, it is critical for participatory research methods to be introduced throughout the research system.