Background: In COVID-19, higher than expected level of intrapulmonary shunt has been described, in association with a discrepancy between the initial relatively preserved lung mechanics and the hypoxia severity. This study aim was to measure the shunt fraction and variations of PaO2/FiO2 ratio and oxygen alveolar-arterial gradient (A-a O2) at different FiO2.Methods: Shunt was measured by a non-invasive system during spontaneous breathing in 12 patients hospitalized at COVID-19 Semi-Intensive Care Unit of Papa Giovanni XXIII Hospital, Bergamo, Italy, between October 22 and November 23, 2020.Results: Nine patients were men, mean age (±SD) 62±15 years, mean BMI 27.5±4.8 Kg/m2. Systemic hypertension, diabetes type 2 and previous myocardial infraction were referred in 33%, 17%, and 7%, respectively. Mean PaO2/FiO2 ratio was 234±66 and 11 patients presented a bilateral chest X-ray involvement. Mean shunt was 21±6%. Mainly in patients with a more severe respiratory failure, we found a progressive decrease of PaO2/FiO2 ratio with higher FiO2. Considering (A-a O2), we found a uniform tendency to increase with FiO2 increasing. Even in this case, the more severe were the patients, the higher was the slope, suggesting FiO2 insensitiveness due to a shunt effect, as strengthened by our measurements.Conclusion: Relying on a single evaluation of PaO2/FiO2 ratio, especially at high FiO2, could be misleading in COVID-19. We propose a two steps evaluation, the first at low SpO2 value (e.g., 92-94%) and the second one at high FiO2 (i.e., >0.7), allowing to characterize both the amendable (ventilation/perfusion mismatch), and the fixed (shunt) contribution quote of respiratory impairment, respectively.