Drawing on advances in nonstationary frequency analysis and the science of causation and attribution, this study employs a nonstationary stochastic paired watershed approach to determine the effect of forest harvesting on floods. Moreover, this study furthers the application of stochastic physics to evaluate the environmental controls and drivers of flood response. Physically-based climate and time-varying harvesting data are used as covariates to drive the nonstationary flood frequency distribution parameters to detect, attribute, and quantify the effect of harvesting on floods in two large, snow-dominated watersheds. Harvesting only 21% of the watershed caused a 38% and 84% increase in the mean but no increase in variability of the frequency distribution in the Deadman River (878 km2) and Joe Ross Creek (99 km2) watersheds, respectively. Consequently, the 7-year, 50-year, and 100-year flood events became approximately two, six, and ten times more frequent in both watersheds. An increase in the mean is posited to occur from an increase in moisture availability following harvest. Variability was not increased because snowmelt synchronization was inhibited by the buffering capacity of abundant lakes, evenly distributed aspects, and widespread spatial distribution of cutblocks in the watersheds. It is the inherent nature of the flood frequency distribution of snowmelt-driven flood regimes which cause even modest increases in magnitude, especially in the upper tail of the distribution, to translate into surprisingly large changes in frequency. Contrary to conventional wisdom, harvesting influenced all flood event sizes, and the sensitivity to harvest increased with increasing flood event size and watershed area.