The CO2 production (rCO2) of eight growing pigs was determined by continuous collection of CO2 over 21 days and simultaneously estimated using the doubly labeled water (DLW) method. The aim was to assess the accuracy of the method before and after correction for known sources of error and to test for any residual discrepancy arising from as yet unidentified sources of error. Mass spectrometer accuracy was verified by analyzing serial dilutions of the dose material in the form of an artificial decay curve; no significant bias was detected. The physiological errors were linearly dependent on weight gain. DLW-derived rCO2 (corrected only for fractionated water loss) underestimated the true value by 0.270 l CO2/g wt gain or -8% in the restricted (group R) and -16% in the ad libitum-fed (group AL) groups. Known sources of error accounted for -0.006 (methane), -0.032 (fecal 2H losses), -0.108 (fat synthesis), and -0.146 (changing pool size) l CO2/g wt gain. After correction for these sources of error the DLW-derived rCO2 differed from the true value by -2 +/- 3% in group R and 0 +/- 3% in group AL. Thus there was no significant bias in the DLW method after correction for known sources of error, even during rapid weight gain or at weight stability with or without correction. The precision estimates include both dose and background errors and uncertainty in the correction factors used. Strategies for optimizing precision are presented.