The National Centers for Environmental Prediction (NCEP) and the University of Washington (UW) run parallel drought monitoring systems over the continental U.S. based on the North American Land Data Assimilation System (NLDAS). Both systems use a suite of land surface models (LSMs), one of which is the Variable Infiltration Capacity (VIC) model. We performed an assessment of differences in drought characteristics estimated using both systems for the period 1979-2006. For soil moisture (SM) percentiles and runoff indices, differences are relatively small among different LSMs in the same system. However, the ensemble mean differences between the two systems are large over the western United States – in some cases exceeding 20% for SM and runoff percentile differences. These differences are most apparent after 2002 when both systems use real-time, rather than retrospective LSM forcings. We performed experiments to determine the sources of these uncertainties. We found that precipitation forcing differences are the primary source of the SM and runoff differences. While temperature and wind speed forcing differences are also large after 2002, their contributions to SM and runoff differences are much smaller than the precipitation contribution.