This study evaluates the performance and internal structure of the distributed hydrology soil vegetation model (DHSVM) using 1998--2001 data collected at Upper Penticton Creek, British Columbia, Canada. It is shown that clear-cut snowmelt rates calculated using data-derived snow albedo curves are in agreement with observed lysimeter outflow. Measurements in a forest stand with 50% air crown closure suggest that the fraction of shortwave radiation transmitted through the canopy is 0.18--0.28 while the hemispherical canopy view factor controlling longwave radiation fluxes to the forest snowpack is estimated at 0.81 ¿ 0.07. DHSVM overestimates shortwave transmittance (0.50) and underestimates the view factor (0.50). An alternative forest radiation balance is formulated that is consistent with the measurements. This new formulation improves model efficiency in simulating streamflow from 0.84 to 0.91 due to greater early season melt that results from the enhanced importance of longwave radiation below the canopy. The model captures differences in canopy rainfall interception between small and large storms, tree transpiration measured over a 6-day summer period, and differences in soil moisture between a dry and a wet summer. While the model was calibrated to 1999 snow water equivalent (SWE) and hydrograph data for the untreated control basin, it successfully simulates forest and clear-cut SWE and streamflow for the 3 other years and 4 years of preharvesting and postharvesting streamflow for the second basin. Comparison of model states with the large array of observations suggests that the modified model provides a reliable tool for assessing forest management impacts in the region.