As human activity and climate variability alter the movement of water through the environment the need to better understand hydrologic cycle responses to these changes has grown. A reasonable starting point for gaining such insight is studying changes in streamflow given the importance of streamflow as a source of renewable freshwater. Using a wavelet assisted method we analyzed trends in the magnitude of annual scale streamflow variability from 967 watersheds in the continental U.S. (CONUS) over a 70 year period (1940–2009). Decreased annual variability was the dominant pattern at the CONUS scale. Ecoregion scale results agreed with the CONUS pattern with the exception of two ecoregions closely divided between increases and decreases and one where increases dominated. A comparison of trends in reference and non-reference watersheds indicated that trend magnitudes in non-reference watersheds were significantly larger than those in reference watersheds. Boosted regression tree (BRT) models were used to study the relationship between watershed characteristics and the magnitude of trends in streamflow. At the CONUS scale, the balance between precipitation and evaporative demand, and measures of geographic location were of high relative importance. Relationships between the magnitude of trends and watershed characteristics at the ecoregion scale exhibited differences from the CONUS results and substantial variability was observed among ecoregions. Additionally, the methodology used here has the potential to serve as a robust framework for top-down, data driven analyses of the relationships between changes in the hydrologic cycle and the spatial context within which those changes occur.