Theoretical and empirical models used to estimate the long‐term water balance of small catchments are inadequate as a check on the observed water balance. Under rigorous research conditions, the observational data are accurate to about ± 13–30% of yield, while the models are accurate only to about ± 10–60% of yield. Analysis of watershed data shows that the uncertainty in any comparison of theoretical and observed values usually is much greater than the difference between the values compared. In general, such comparisons are not helpful either in confirming or challenging the accuracy of water balance measurements, or in assessing the watertightness of catchment areas.