The pattern of streamflow recession after rain events offers clues about the relationship between watershed runoff (observable as river discharge) and water storage (not directly observable) and can help in water resource assessment and prediction. However, there have been few systematic assessments of how streamflow recession varies across flow rates and how it relates to independent assessments of terrestrial water storage. We characterized the streamflow recession pattern in 61 relatively undisturbed small watersheds (1–100 km2 ) across the coterminous United States with multiyear records of hourly streamflow from automated gauges. We used the North American Regional Reanalysis to help identify periods where precipitation, snowmelt, and evaporation were small compared to streamflow. The order of magnitude of the recession timescale increases from 1 day at high flow rates (∼1 mm h−1 ) to 10 days at low flow rates (∼0.01 mm h−1 ), leveling off at low flow rates. There is significant variability in the recession timescale at a given flow rate between basins, which correlates with climate and geomorphic variables such as the ratio of mean streamflow to precipitation and soil water infiltration capacity. Stepwise multiple regression was used to construct a six-variable predictive model that explained some 80 % of the variance in recession timescale at high flow rates and 30–50 % at low flow rates. Seasonal and interannual variability in inferred storage shows similar time evolution to regional-scale water storage variability estimated from GRACE satellite gravity data and from land surface modeling forced by observed meteorology, but is up to a factor of 10 smaller. Study of this discrepancy in the inferred storage amplitude may provide clues to the range of validity of the recession curve approach to relating runoff and storage.