We describe an experimental end-to-end streamflow forecasting approach that uses NCEP Global Spectral Model (GSM) ensemble climate forecasts to drive the Variable Infiltration Capacity (VIC) macroscale hydrologic model for large river basins. The method was initially tested over the East Coast U.S. during the summer 2000 drought, and subsequently over a western U.S. domain that includes the Columbia, Colorado, and Sacramento-San Joaquin Rivers, beginning in March 2001. The linkage of the climate and hydrologic models provides a mechanism for exploiting potential skill that results from either (or the combination of) climate forecasts, or knowledge of hydrologic initial conditions (soil moisture, snow), with the intent of improving streamflow forecasts at lead times of up to six months. Each month, GSM precipitation and temperature ensemble forecasts are adjusted to remove climate model bias, and the bias-adjusted ensembles are then used to drive the hydrologic model, which in turn produces ensemble streamflow forecasts with six-month lead times. Initial hydrologic conditions for the forecasts are estimated by driving the VIC model with observed meteorological data (gridded product derived from NCDC cooperator stations) and updated for the most recent three months using archived real-time Land Data Assimilation System (LDAS) gridded forcings. In the regional applications, the climate model forecasts mostly tended toward climate normals during the test periods, and anomalous initial hydrologic states dominated most of the hydrologic forecasts. In the East Coast application, critically low soil moisture in the spring of 2000 persisted into summer, despite a return to more normal atmospheric conditions during the summer. A retrospective analysis of a year with significant climate forecast anomalies over the western U.S. (the 1997-98 ENSO event), however, showed that streamflow prediction skill can result from the climate forecast signal as well. In the Pacific Northwest, the summer 2001 forecast simulations predicted, among other effects, the severe deficit in snowpack, runoff and soil moisture; hence summer streamflow, starting in March. In this case, prediction skill appeared to derive primarily from hydrologic persistence associated with the initial states of snowpack during the spring.