We validated doubly-labelled water (DLW) by comparison to indirect calorimetry and food intake–mass balance in eight Labrador dogs (24–32 kg) over 4 d. We used several alternative equations for calculating CO2 production, based on the single- and two-pool models and used two alternative methods for evaluating the elimination constants: two-sample and multiple-sampling. In all cases the DLW technique overestimated the direct estimate of CO2 production. The greatest overestimates occurred with the single-pool model. Using two samples, rather than multiple samples, to derive the elimination constants produced slightly more discrepant results. Discrepancies greatly exceeded the measured analytical precision of the DLW estimates. The higher values with DLW probably occurred because the dogs were extremely active during the 1 h in each 24 spent outside the chamber. Estimates of CO2 production from food intake–mass balance, which include this activity, produced a much closer comparison to DLW (lowest mean discrepancy 0·3 % using the observed group mean dilution space ratio and an assumption that the mass changes reflected changes in hydration for all except one animal). We recommend an equilibration time of 6 h and use of the two-pool model based on the observed population dilution space for future studies of energy demands in dogs of this body mass.