Background & aims: Both overfeeding and underfeeding of intensive care unit (ICU) patients are associated with worse outcomes. Predictive equations of nutritional requirements, though easily implemented, are highly inaccurate. Ideally, the individual caloric target is based on the frequent assessment of energy expenditure (EE). Indirect calorimetry is considered the gold standard but is not always available. EE estimated by ventilator-derived carbon dioxide consumption (EEVCO 2) has been proposed as an alternative to indirect calorimetry, but there is limited evidence to support the use of this method. Methods: We prospectively studied a cohort of adult critically ill patients requiring mechanical ventilation and artificial nutrition. We aimed to compare the performance of the EEVCO 2 with the EE measured by indirect calorimetry through the calculation of bias and precision (accuracy), agreement, reliability and 10% accuracy rates. The effect of including the food quotient (nutrition intake derived respiratory quotient) in contrast to a fixed respiratory quotient (0.86), into the EEVCO 2 formula was also evaluated. Results: In 31 mechanically ventilated patients, a total of 414 paired measurements were obtained. The mean estimated EEVCO 2 was 2134 kcal/24 h, and the mean estimated EE by indirect calorimetry was 1623 kcal/24 h, depicting a significant bias of 511 kcal (95% CI 467e560, p < 0.001). The precision of EEVCO 2 was low (lower and upper limit of agreement À63.1 kcal and 1087. o kcal), the reliability was good (intraclass correlation coefficient 0.613; 95% CI 0.550e0.669, p < 0.001) and the 10% accuracy rate was 7.0%. The food quotient was not significantly different from the respiratory quotient (0.870 vs. 0.878), with a small bias of 0.007 (95% CI 0.000e0.015, p ¼ 0.54), low precision (lower and upper limit of agreement À0.16 and 0.13), poor reliability (intraclass correlation coefficient 0.148; 95% CI 0.053e0.240, p ¼ 0.001) and a 10% accuracy rate of 77.5%. Estimated mean EEVCO 2 , including the food quotient, was 2120 kcal/24 h, with a significant bias of 496 kcal (95% CI 451e542; p < 0.001) and low precision (lower and upper limit of agreement À157.6 kcal and 1170.3 kcal). The reliability with EE estimated by indirect calorimetry was good (intraclass correlation coefficient 0.610, 95% CI 0.550e0.661, p < 0.001), and the 10% accuracy rate was 9.2%. Conclusions: EEVCO 2 , compared with indirect calorimetry, overestimates actual energy expenditure. Although the reliability is acceptable, bias is significant, and the precision and accuracy rates are unacceptably low when the VCO 2 method is used. Including the food quotient into the EEVCO 2 equation does not improve its performance. Predictive equations, although inaccurate, may even predict energy expenditure better compared with the VCO 2-method. Indirect calorimetry remains the gold standard method.