Complete sampling of expired air is essential for accurate O2 consumption(CO2 production) [VO2(VCO2)] measurements with flow-through indirect calorimetry. In preterm infants complete sampling is critical, because only low sampling flows can be used. The accuracy of the various breath sampling systems at low flows and their patient compatibility is untested. We therefore measured 1) the accuracy of VO2(VCO2) measurements with a face mask, a head hood, and a canopy in vitro at low sampling flows; 2) the effect of breathing on measurements with the face mask; and 3) the effect of breath sampling systems on activity and body temperature of preterm infants. VO2(VCO2) were measured with a Deltatrac II. In vitro we used a methanol miniburner incorporated into a doll, which could simulate low VO2(VCO2) and tidal breathing. In vivo we studied seven preterm infants < 1500 g. With the face mask VO2(VCO2) measurements were accurate at a flow of 3 L/min (error -1 +/- 0.8%), when tidal volume was < 15 mL/breath and the distance between mask and manikin < 1 cm. With hood and canopy VO2(VCO2) were underestimated at a flow of 3 L/min (error -13 +/- 1% and -14 +/- 5%), and results were markedly influenced by body position. For accurate measurements, the hood needed a flow of 4.5 L/min, the canopy 8.3 L/min. In vivo the face mask did not increase heart rate, respiration, activity, or rectal temperature, but hood and canopy increased rectal temperature by 0.3-0.4 degree C. For VO2(VCO2) measurements in infants < 1500 g, a face mask should be used, which is accurate at low flows and does not change body temperature. Accuracy at low flows and patient compatibility of breath sampling systems should be evaluated and reported for VO2(VCO2) measurements in preterm infants.