Bracketing the uncertainty of streamflow and agricultural runoff under climate change is critical for proper future water resource management in agricultural watersheds. This study used the Soil and Water Assessment Tool (SWAT) in conjunction with a Latin hypercube climate change sampling algorithm to construct a 95% confidence interval (95CI) around streamflow, sediment load, and nitrate load predictions under changes in climate for the Sacramento and San Joaquin River watersheds in California's Central Valley. The Latin hypercube algorithm sampled 2000 combinations of precipitation and temperature changes based on Intergovernmental Panel on Climate Change projections from multiple General Circulation Models. Average monthly percent changes of the upper and lower 95CI limits compared to the present-day simulation and a statistic termed the “r-factor” (average width of the 95CI band divided by the standard deviation of the 95CI bandwidth) were used to assess watershed sensitivities. 95CI results indicate that streamflow and sediment runoff in the Sacramento River watershed are more likely to decrease under climate change compared to present-day conditions, whereas the increase and decrease for nitrate runoff were found to be equal. For the San Joaquin River watershed, streamflow slightly decreased under climate change, whereas sediment and nitrate runoff increased compared to present-day climate. Comparisons of watershed sensitivities indicate that the San Joaquin River watershed is more sensitive to climate changes than the Sacramento River watershed, which is largely caused by the high density of agricultural land. Copyright © 2012 John Wiley & Sons, Ltd.