Data from the 1994 USDA nationwide survey (CSFII) on 190 non-smoking males (aged 20-29) were used to propose a method for adjusting total water intake for the diuretic effects of caffeine and alcohol, and evaluate the potential for related misclassification bias. The data were processed on a per meal basis. Under the assumption that subjects were in water balance at the start of the survey day, water losses due to caffeine (1.17 ml/mg caffeine) and alcohol (10 ml/g alcohol) were subtracted from crude intake estimates. If water intake for one meal was inadequate for excretion of the associated osmotic load at 750 mosmol/l, water losses for the subsequent meal were reduced by 32%. Unadjusted and adjusted mean total water intakes differed by 321.5 g. Misclassification appeared worst at higher water intakes. Linear regression models, each with a water intake variable as an independent variable and body mass index as the outcome, were fit to evaluate the potential for alcohol- and caffeine-related misclassification bias. Misclassification resulted in large changes (all >10%) in linear regression estimates of effect. Future studies of water-disease relationships, especially those intending to compare extremes of total water intake, should consider caffeine- and alcohol-related misclassification bias.