River restoration is at the forefront of applied hydrologic science. However, many river restoration projects are conducted with minimal scientific context. We propose two themes around which a research agenda to advance the scientific basis for river restoration can be built. First, because natural variability is an inherent feature of all river systems, we hypothesize that restoration of process is more likely to succeed than restoration aimed at a fixed end point. Second, because physical, chemical, and biological processes are interconnected in complex ways across watersheds and across timescales, we hypothesize that restoration projects are more likely to be successful in achieving goals if undertaken in the context of entire watersheds. To achieve restoration objectives, the science of river restoration must include (1) an explicit recognition of the known complexities and uncertainties, (2) continued development of a theoretical framework that enables us to identify generalities among river systems and to ask relevant questions, (3) enhancing the science and use of restoration monitoring by measuring the most effective set of variables at the correct scales of measurement, (4) linking science and implementation, and (5) developing methods of restoration that are effective within existing constraints. Key limitations to river restoration include a lack of scientific knowledge of watershed-scale process dynamics, institutional structures that are poorly suited to large-scale adaptive management, and a lack of political support to reestablish delivery of the ecosystem amenities lost through river degradation. This paper outlines an approach for addressing these shortcomings.