Among the many natural disasters that disrupt human and industrial activity in the United States each year, including tornadoes, hurricanes, extreme temperatures, and lightning, floods are among the most devastating and rank second in the loss of life. Indeed, the societal impact of floods has increased during the past few years and shows no sign of abating. Although the scientific questions associated with flooding and its accurate prediction are many and complex, an unprecedented opportunity now exists - in light of new observational and computing systems and infrastructures, a much improved understanding of small-scale meteorological and hydrological processes, and the availability of sophisticated numerical models and data assimilation systems - to attack the flood forecasting problem in a comprehensive manner that will yield significant new scientific insights and corresponding practical benefits. The authors present herein a set of recommendations for advancing our understanding of floods via the creation of natural laboratories situated in a variety of local meteorological and hydrological settings. Emphasis is given to floods caused by convection and cold season events, fronts and extratropical cyclones, orographic forcing, and hurricanes and tropical cyclones following landfall. Although the particular research strategies applied within each laboratory setting will necessarily vary, all will share the following principal elements: (a) exploitation of those couplings important to flooding that exist between meteorological and hydrological processes and models; (b) innovative use of operational radars, research radars, satellites, and rain gauges to provide detailed spatial characterizations of precipitation fields and rates, along with the use of this information in hydrological models and for improving and validating microphysical algorithms in meteorological models; (c) comparisons of quantitative precipitation estimation algorithms from both research (especially multiparameter) and operational radars against gauge data as well as output produced by meso- and storm-scale models; (d) use of data from dense, temporary river gauge networks to trace the fate of rain from its starting location in small basins to the entire stream and river network; and (e) sensitivity testing in the design and implementation of separate as well as coupled meteorological and hydrologic models, the latter designed to better represent those nonlinear feedbacks between the atmosphere and land that are known to play an important role in runoff prediction. Vital to this effort will be the creation of effective and sustained linkages between the historically separate though scientifically related disciplines of meteorology and hydrology, as well as their observational infrastructures and research methodologies.