The relationship between total body water (TBW) and extracellular water (ECW), measured by deuterium oxide dilution and bromide dilution respectively, and impedance and impedance index (height2/impedance) at 1, 5, 50 and 100 kHz was studied. After correction for TBW, ECW was correlated only with the impedance index at 1 and 5 kHz. After correction for ECW, TBW was best correlated with the impedance index at 100 kHz. The correlation of body-water compartments with impedance values obtained with modelling programs was lower than with measured impedance values. Prediction formulas for ECW (at 1 and 5 kHz) and TBW (at 50 and 100 kHz) were developed. The prediction errors for ECW and TBW were 1.0 and 1.7 kg respectively (coefficient of variation 5%). The residuals of both ECW and TBW were related to the ECW/TBW value. Application of the prediction formulas in a population, independently measured, revealed a slight overestimation of TBW and ECW, which could be largely explained by differences in the validation group in body-water distribution and in body build. The ratio of impedance at 1 kHz to impedance at 100 kHz was correlated with body-water distribution (ECW/TBW). The relation is however not strong enough to be useful as a predictor. It is concluded that an independent prediction of ECW and TBW, using impedance at low and high frequency respectively, is possible, but that the bias depends on the body-water distribution and body build of the measured subject.