The effect of PEEP on the arterial minus end-tidal carbon dioxide gradient (PaCO2-PetCO2) was evaluated in 13 adult patients with acute respiratory failure. The morphologic study of the pressure-volume (P-V) curves allowed separation of the patients into two groups: group 1 (n = 7) with initial inflection point in the (P-V) curve, and group 2 without inflection point. We hypothesized that the profile of the PaCO2-PetCO2 gradient would indicate an appropriate PEEP level only in patients with recruitable air spaces. We ventilated group 1 patients with zero end expiratory pressure (ZEEP), PEEP corresponding to inflection point pressure (PEEPPi) and PEEP5 cm H2O above PEEPPi, and group 2 patients with ZEEP, 6 cm H2O PEEP and 12 cm H2O PEEP. The PaCO2-PetCO2 gradient changed significantly in group 1 (ZEEP: 13.59 mm Hg; PEEPPi: 8.33 mm Hg; PEEPPi + 5 cm H2O: 10.54 mm Hg), but not in group 2 (ZEEP: 14.15 mm Hg; PEEP 6 cm H2O: 14.20 mm Hg; PEEP 12 cm H2O: 16.53 mm Hg). Our results show that the PaCO2-PetCO2 gradient may be useful in selecting a PEEP level which produces alveolar recruitment, but only in those patients with initial inflection point in the P-V curve.