PURPOSE: To determine whether fever is associated with an increased or decreased risk of death in patients admitted to an intensive care unit (ICU) with infection. METHODS: We evaluated the independent association between peak temperature in the first 24 h after ICU admission and in-hospital mortality according to whether there was an admission diagnosis of infection using a database of admissions to 129 ICUs in Australia and New Zealand (ANZ) (n = 269,078). Subsequently, we sought to confirm or refute the ANZ database findings using a validation cohort of admissions to 201 ICUs in the UK (n = 366,973). RESULTS: A total of 29,083/269,078 (10.8%) ANZ patients and 103,191/366,973 (28.1%) of UK patients were categorised as having an infection. In the ANZ cohort, adjusted in-hospital mortality risk progressively decreased with increasing peak temperature in patients with infection. Relative to the risk at 36.5-36.9°C, the lowest risk was at 39-39.4°C (adjusted OR 0.56; 95% CI 0.48-0.66). In patients without infection, the adjusted mortality risk progressively increased above 39.0°C (adjusted OR 2.07 at 40.0°C or above; 95% CI 1.68-2.55). In the UK cohort, findings were similar with adjusted odds ratios at corresponding temperatures of 0.77 (95% CI 0.71-0.85) and 1.94 (95% CI 1.60-2.34) for infection and non-infection groups, respectively. CONCLUSIONS: Elevated peak temperature in the first 24 h in ICU is associated with decreased in-hospital mortality in critically ill patients with an infection; randomised trials are needed to determine whether controlling fever increases mortality in such patients.