Oxygen therapy is used to reverse hypoxemia since more than a century. Current usage is broader and includes routine oxygen administration despite normoxemia which may result in prolonged periods of hyperoxemia. While systematic oxygen therapy was expected to be of benefit in some ischemic diseases such as stroke or acute myocardial infarction, recent randomised controlled trials (RCTs) have challenged this hypothesis by showing the absence of clinical improvement. Although oxygen is known to be toxic at high inspired oxygen fractions, a recent meta-analysis of RCTs revealed the life-threatening effect of hyperoxemia, with a dose-dependent relationship. Several recommendations have therefore been updated: (i) to monitor peripheral oxygen saturation (SpO2) as a surrogate for arterial oxygen saturation (SaO2); (ii) to initiate oxygen only when the lower SpO2 threshold is crossed; (iii) to titrate the delivered oxygen fraction to maintain SpO2 within a target range; and (iv) to stop supplying oxygen when the upper limit of SpO2 is surpassed, in order to prevent hyperoxemia. The lower and upper limits of SpO2 depend on the presence of risk factors for oxygen-induced hypercapnia (Chronic obstructive pulmonary disease, asthma, and obesity-associated hypoventilation). For patients at risk, oxygen therapy should be started when SpO2 is≤88% and stopped when it is>92%. For patients without risk factors, oxygen therapy should be started when SpO2 is≤92% and stopped when it is >96%. High-flow oxygen should only be used in a few diseases such as carbon monoxide poisoning, cluster headaches, sickle cell crisis and pneumothorax. Copyright © 2019 Société Nationale Française de Médecine Interne (SNFMI). Published by Elsevier Masson SAS. All rights reserved.