Estimates of body composition by the deuterium dilution technique were made in 55 healthy subjects (38 men and 17 women) with a mean age of 28.4 years, weight 65.5 kg and height 173.7 cm (body mass index 16.9-29.4 kg/m2). The results were compared with estimates obtained in the same subjects by 5 bedside techniques (skinfold thickness, impedance, resistance and 2 equations predicting body composition from weight and height). The results of all the bedside predictive methods were found to differ significantly from those of the deuterium dilution method (P less than 0.001). The Holtain impedance method underestimated the fat-free mass (FFM) obtained by deuterium dilution (50.8 +/- 7.9 kg) by a mean difference (mean bias) of 4.1 kg while the other methods overestimated FFM (mean bias or mean difference of -1.3 to -2.4 kg). The resistance (Valhalla equations) and skinfold methods showed the narrowest 95 per cent limits of agreement, when compared with the deuterium dilution technique, while the weight and height equations showed the widest limits of agreement. It is concluded that in this population, the resistance (Valhalla equations) and the skinfold thickness methods were the best predictors of body composition as measured by deuterium dilution. Discrepancies between the bedside methods and deuterium dilution method may be due to (a) methodological or biological differences between the test population used in this study and the original population used to validate the method, (b) differences in the structure of the equations used to interpret the measured biological variables, such as impedance and resistance, and (c) the use of predictive equations that are based on different 'reference' methods, such as deuterium dilution or densitometry, and on different assumptions. Predictive equations based on a combination of these reference methods are preferable to those based on single methods.