Switchgrass (Panicum virgatum L.) has attracted attention as a promising second generation biofuel feedstock. Both existing grasslands and marginal croplands have been suggested as targets for conversion to switchgrass, but the resulting production potentials and hydrologic impacts are not clear. The objectives of this study were to model switchgrass production on existing grasslands (scenario-I) and on marginal croplands that have severe to very severe limitations for crop production (scenario-II) and to evaluate the effects on evapotranspiration (ET) and streamflow. The Soil and Water Assessment Tool (SWAT) was applied to the 1063 km2 Skeleton Creek watershed in north-central Oklahoma, a watershed dominated by grasslands (35%) and winter wheat cropland (47%). The simulated average annual yield (2002-2011) for rainfed Alamo switchgrass for both scenarios was 12 Mg ha-1. Yield varied spatially under scenario-I from 6.1 to 15.3 Mg ha-1, while under scenario-II the range was from 8.2 to 13.8 Mg ha-1. Comparison of average annual ET and streamflow between the baseline simulation (existing land use) and scenario-I showed that scenario-I had 5.6% (37 mm) higher average annual ET and 27.7% lower streamflow, representing a 40.7 million m3 yr-1 streamflow reduction. Compared to the baseline, scenario-II had only 0.5% higher ET and 3.2% lower streamflow, but some monthly impacts were larger. In this watershed, the water yield reduction per ton of biomass production (i.e. hydrologic cost-effectiveness ratio) was more than 5X greater under scenario-I than under scenario-II. These results suggest that, from a hydrologic perspective, it may be preferable to convert marginal cropland to switchgrass production rather than converting existing grasslands.