1. In order to establish biological and/or methodological explanations for the wide variability in recovery (50-100%) of labelled CO2 after administration of [13C]bicarbonate or [14C]bicarbonate, 34 human bicarbonate studies involving 480 subjects were analysed, and potential methodological issues were investigated in the laboratory. 2. Overall, continuous infusion studies reported a higher recovery than bolus studies (84 +/- 11% versus 69 +/- 12%; P < 0.001). No significant differences in recovery were found between 14C and 13C studies, children and adults, obese and lean subjects, or rest and exercise (steady state). Higher recoveries were found during feeding than during fasting (84 +/- 8% versus 74 +/- 7%; P < 0.001). Different methods used to analyse the results (0-10%) and different study protocols, which include differences in the duration of infusions and background drift in 13C enrichment (0-10%), contribute to the variability. 3. The laboratory studies suggest multiple sources of potential error, including loss of CO2 from the scintillation fluid (up to > 30%, but only in 14C studies in which the scintillation fluid is not alkalized), diffusion of CO2 through syringes and tubing (0 to > 10%), non-linearity of CO2 analysers (up to 8%), inaccuracies in the measurement of bicarbonate concentrations (13C studies) or the strength of CO2-trapping agents (14C studies; 0-8%). 4. It is concluded that much of the variability in the recovery of labelled bicarbonate is likely to be attributable to methodological differences, and that attention to these will ensure better interpretation of metabolic studies that involve oxidation of carbon-labelled substrates.