BACKGROUND Supplemental oxygen therapy is used for intensive care (ICU) patients with severe sepsis, but with no general guidelines and few safety data. The aim of this observational study was to describe the variability in oxygen administration as well as the association between partial pressure of arterial oxygen (PaO2 ) and mortality. METHODS We extracted data from two Scandinavian clinical trials of ICU patients with severe sepsis or septic shock. We calculated average PaO2 and fraction of inspired oxygen (FiO2 ) from trial inclusion and the following 5 days, and assessed the association between PaO2 and 90-day mortality. RESULTS The median PaO2 was 9.8 kPa [5-95% range 6.4-19.9] and FiO2 was 0.51 [5-95% range 0.27-1.00], respectively. Eight hundred and five of 1,770 patients (45%) died. The relative risk of mortality was 1.43 [95% CI: 1.19-1.65] in patients with average PaO2 < 8 kPa and 1.29 [95% CI: 0.84-1.68] in patients with average PaO2 ≥ 16 kPa, as compared to patients with average PaO2 10-12 kPa. The relative risk of mortality was 1.38 [95% CI: 1.17-1.58] in patients with an average FiO2 0.60-0.80 and 2.10 [95% CI: 1.88-2.23] in patients with an average FiO2 ≥ 0.80 as compared to patients with an average FiO2 ≤ 0.40. CONCLUSION Administration of oxygen in patients with severe sepsis resulted in a wide range of PaO2 . Significantly higher mortality was observed in patients with an average PaO2 < 8 kPa and FiO2 ≥ 0.60. The results do not imply causation and the associations between average PaO2 and adverse outcomes have to be assessed further.