This study assessed the validity of bioelectrical impedance analysis (BIA) to predict total body water (TBW) in patients with cirrhosis. TBW was estimated by deuterium oxide dilution (D2O) and compared with TBW predicted by BIA in 27 patients with cirrhosis with and without ascites or edema (Group A), in a subgroup of 18 'dry' cirrhotics without clinical signs of fluid overload (Group B) and in 27 healthy controls. Three different BIA regression equations were used. In all three groups of subjects high correlation coefficients were obtained between D2O-TBW and BIA-TBW (r > 0.88). In the cirrhotic Group A, BIA significantly underpredicted D2O-TBW by all 3 equations (2.9-3.8 l) and the regression lines were different from the lines of identity by two equations. Standard errors of estimate were high in Group A (3.04-3.97 l) in comparison with Group B (1.79-2.46 l) and the controls (1.03-1.41 l). In the 'dry' cirrhotics (Group B) and in the controls, TBW was correctly predicted by two of three BIA equations, and regression lines were not significantly different from lines of identity. Correlation coefficients in Group B were higher than in Group A (r = 0.96-0.97 vs. 0.89-0.92) and were comparable with controls (r = 0.98-0.99). We conclude that BIA is not a valid method of estimating TBW in cirrhotic patients with ascites and edema. In cirrhotic patients without clinical signs of fluid overload BIA can be used to predict TBW, although accuracy is slightly lower than in healthy controls.