In the doubly-labelled water (2H2(18)O) method for the measurement of carbon dioxide production rate in man, single exponential disappearance curves for 2H2O and H2(18)O in body water are used; the precision with which the slopes and intercepts of the curves are estimated determines the precision of the estimate of carbon dioxide production rate. In studies with infants, and in computer simulations, the effect of different experimental regimes on the overall precision of the carbon dioxide production estimate was investigated. When the number of data points used was progressively reduced by shortening the total observation period from 7 d (about 3 biological half-lives for the isotopes) to 1 d there was a deleterious effect on precision and in the infants there was an upward bias in the values for carbon dioxide production. When the number of data points was reduced by removing points from the middle of the exponential curves with the maintenance of the 7-d experimental period, precision was also reduced but by less than in the former procedure and there were no consistent trends in the average values for carbon dioxide production.