To examine the effect of auto-positive end-expiratory pressure (autoPEEP) on the estimation of arterial carbon dioxide pressure (PaCO2) from end-tidal carbon dioxide pressure (PetCO2) during changes in minute ventilation (MV), we studied 24 consecutive sedated and paralyzed patients under controlled mechanical ventilation for acute respiratory failure. The patients were grouped according to whether they had autoPEEP: group I (n = 11) comprised non-autoPEEP patients and group II (n = 13) comprised autoPEEP patients. Patients were randomly ventilated at three different levels of MV: normal MV (basal tidal volume), high MV (tidal volume 2.5 mL/kg above basal), and low MV (tidal volume 2.5 mL/ kg below basal). Respiratory rate and inspiration to expiration ratio were kept constant during the study. In each condition, we measured arterial blood gases, expiratory capnograms, airway pressure, and autoPEEP. We determined PaCO2-PetCO2 gradient, predicted PaCo2 (Pa'CO2) [Pa'CO2 = PetCO2 for each condition + (PaCO2 PetCO2 gradient at normal MV)], and expired CO2 slope. The PaCO2-PetCO2 gradient only remained stable in group I (mean values for low, normal, and high MV were 3.3, 3.3, and 3.5 mm Hg, respectively), while group II showed a significant difference during low MV (12.2 mm Hg) when compared with normal MV (8.4 mm Hg; P < .01) and high MV (8.9 mm Hg; P < .05). PaCO2 and PetCO2 showed significant correlations in both groups (r = .92 in group I and .79 in group II). However, Pa'CO2 could only be safely estimated in patients without autoPEEP when the difference between PaCO2 and Pa'CO2, ranged between 1.6 and -1.9 mm Hg. Slopes of expired CO2 greater than 3 mm Hg/s identified patients with autoPEEP of 89% sensitivity, 93% specificity, 94% positive predictive power, and 95% accuracy. A significant correlation was found between autoPEEP and expired CO2 slope (r = .70; P < .001), between autoPEEP and PaCO2-PetCO2 gradient (r = .46; P < .001), and between CO2 expired slope and PaCO2 PetCO2 gradient (r = .74; P < .001). These results indicate that in patients with acute respiratory failure under controlled mechanical ventilation, the presence of autoPEEP is associated with inaccuracy in the calculation of predicted PacoZ from PetcoZ after changes in MV at fixed respiratory rates. © 1991.