Hydrologic prediction errors arise from uncertainty in initial moisture states (mainly snowpack and soil moisture), in boundary forcings (primarily future precipitation and temperature), and from model structure and parameter uncertainty. We evaluate the relative importance of initial condition and boundary forcing uncertainties using a hindcast-based framework that contrasts Ensemble Streamflow Prediction (ESP) with an approach that we term “reverse-ESP”. In ESP, a hydrologic model with assumed perfect initial conditions (ICs) is forced by a forecast ensemble resampled from observed meteorological sequences; whereas reverse-ESP combines an ensemble of resampled ICs with a perfect meteorological forecast. The framework shows that in northern California, US, ICs yield streamflow prediction skill for up to 5 months during the transition between the wet and dry seasons, whereas during the reverse transition, climate forecast information is critical. In southern Colorado, IC knowledge outweighs climate prediction skill for shorter periods due to a more uniform precipitation regime.