Analysis of the mean annual cycle of the observed water budget based on radiosonde data for the 20 year period 1973--1992 reveals a negative bias in the atmospheric estimate of mean moisture flux convergence over the central United States when compared to mean river discharge for the same period. Several possible sources of this bias are examined through the analysis of radiosonde, wind profiler, and river discharge data as well as water budget estimates from NCEP's mesoscale Eta Model. The analysis suggests that 20 years of radiosonde observations are sufficient to get stable mean water vapor flux convergence estimates. The analysis also shows that the bias between averaged vapor flux convergence and river discharge likely does not depend on changes in the locations of upper-air observations or on changes in the radiosondes observational technique. Components of the atmospheric water budget diurnal cycle are estimated from hourly wind profiler data and from 3-hourly Eta Model output. The analyses show that monthly mean atmospheric water budget quantities calculated from twice a day sampling, on average, significantly underestimate magnitudes of these quantities compared to estimates based on hourly or 3-hourly analyses. The magnitude of the underestimation is more than 60% of the difference between mean observed terrestrial and atmospheric components of the hydrological cycle over the central United States, in agreement with earlier studies suggesting that temporal sampling may account for part of this bias. The effect of spatial sampling was evaluated using differences between atmospheric water budget terms from the Eta Model at its full 48 km resolution compared to Eta Model water budget estimates sampled at the radiosonde spatial resolution. These analyses show that the magnitude of the effect of spatial resolution is about 40% of the magnitude of the mean bias between the average moisture flux convergence and river discharge. The combined effect of temporal and spatial undersampling by radiosondes over the central United States is estimated to be 0.44 mm d-1. This value is within 8% of the estimated mean difference between river discharge and moisture flux convergence for the central United States. ¿ 1999 American Geophysical Union