A lot of initiatives for improving water quality have been developed over the last 15 y in Brittany in response to degradation induced by intensive farming and under the pressure of European policy and environmental organizations. This has involved the partnerships of farmer organizations, organizations in charge of rural affairs, research and formation institutes, and environmental nongovernmental organizations. In this paper, we present 2 complementary aspects of an original, and possibly efficient, water policy within the framework of water management in a medium-sized watershed, including 1) development of new methods of diagnostic and decision support based on participative approaches and 2) development of new methods to assess the current status and effect of alternative scenarios, taking into account the complexity of a system with strong agricultural and hydrological variability and a relatively long response time. The 1st series of methods, which deals with the buffering capacity of landscape structures, is close to a social learning approach; the 2nd illustrates the importance, for policy makers, of a precisely defined protocol for data monitoring and analysis and of the use of spatially distributed and dynamic models when water policy is based on an obligation of results. In spite of the coexistence of all the necessary constituents of a coherent policy, it seems difficult to build. The state of current water quality illustrates the importance and limitations of incentive policy.