Hydrologic forecasts for the Upper Gunnison River Basin (GRB) in southwestern Colorado are produced using the Precipitation Runoff Modeling System (PRMS) as part of the Watershed and River System Management Program. Two forecasting techniques are compared in this study. Fourteen-day forecasts of streamflow produced using Ensemble Streamflow Prediction (ESP) techniques are compared with those produced by Statistically Down Scaling (SDS) hydrologic input from a Medium Range Forecast model. The SDS procedure uses a step-wise multiple linear regression approach to downscale precipitation and temperature to each of the fifteen subbasins within the GRB. Residuals in the regression equation are modeled stochastically to provide 100 ensemble forecasts for each subbasin. A re-ordering method (the Schaake Shuffle) was used to recover the space-time variability in the downscaled precipitation and temperature fields. For each technique, streamflow forecasts from each subbasin were accumulated and routed through the critical management nodes to the outlet of the GRB. Measured and/or reconstructed streamflow was not available for use in determining forecast accuracy at these locations. But, SDS forecasts showed increased skill and reduced spread of the ensemble output when compared with those produced using the ESP technique in the five headwater subbasins with unimpaired measured daily streamflow and the seven subbasins with reconstructed monthly streamflows.