Managing a catchment for drinking water supply with a high proportion of agricultural land use is a difficult task if one has to maintain a reasonable balance between water quality demand and consequent restrictions for the farming industry. In this paper, we present a neural net-based method for finding good approximations to solutions of this problem. This method is capable of ‘inverting’ a hydrological model to identify land use scenarios that match best the leaching criteria defined for establishing a certain water quality level in the stream. The method not only allows simulation land use scenarios like hydrologic models do, but can search systematically for land use scenarios that fulfill specified criteria without worrying about the complexity of combinatorial optimisation.