This paper compares with observations the energy and water budgets for the subbasins of the Mississippi (the Arkansas-Red, the upper Missouri, the upper Mississippi, the Ohio, and the lower Mississippi and Tennessee Rivers), which were computed on-line with an hourly time scale from the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis from 1985 to 1993. The model has a significant precipitation spin-up between the analysis cycle and the 12--24 hour forecast, ranging from 24% to about 40% for the drier Missouri basin. The spin-up of the model large-scale precipitation ranges from 30 to 50%, roughly double that of the spin-up of the model convective precipitation. The model has an erroneous peak in convective precipitation near local noon, but on 2 day and monthly timescales, the 12--24 hour forecast precipitation is only 10 to 20% higher than the observed precipitation for most of the subbasins. The model runoff, which is all deep runoff from the base soil layer, is low on an annual basis, primarily because the model has very little Spring runoff. The nudging of soil water in the analysis cycle, based on 0-6 hour forecast errors in low-level humidity, plays a major role in the model liquid hydrology. The nudging term has a large annual cycle, positive in summer and negative in winter. Although nudging prevents the downward interannual drift of soil water, associated with a shortfall of precipitation in the analysis cycle, it also attempts to compensate for other errors in the model, such as errors in the seasonal cycle of evaporation and runoff, and may damp the variability of soil water. The model frozen hydrology in winter is not conservative and snowmelt is probably too small. Overall, the ECMWF reanalysis gives a valuable description of the surface energy and water balance of the Mississippi River subbasins on timescales longer than the diurnal, and at the same time, it is clear that improvements in the model physics are needed. ¿ 1999 American Geophysical Union