A study of the model-estimated soil water, aggregated over three large drainage basins of the Midwestern USA, is reported. The basin areas are in the range from 2000 km(2) to 3500 km(2), and allow the study of mesoscale (1000-10 000 km(2)) soil water features. In each case, a conceptual hydrologic model was used to produce upper and lower soil water estimates that are consistent with the atmospheric forcing of daily precipitation, potential evapotranspiration and air temperature, and with the observed daily streamflow divergence over a 40 year period. It is shown that the water contents of the upper and lower soil reach peaks in different months, with the soil column being most saturated in June, when the area is prone to serious flooding. Temporal and spatial features of the variability of model-estimated soil water content are identified. The autocorrelation function of monthly averaged soil water shows that the upper soil water remains persistent for about a season, whereas the persistence of the lower soil water extends to several seasons. The soil water estimates of the three study basins exhibit strong similarities in annual cycles and interannual variability. It is shown that the frequency of significant positive (wet) soil water anomalies that extend over a 2 degrees x 2 degrees region is lower than that of significant negative (dry) ones of the same extent in this region of the USA.