Background: Body temperature is a time-honored marker of serious bacterial infection, but there are few studies of its test performance. The aim of our study was to determine the accuracy of temperature measured on presentation to medical care for detecting serious bacterial infection. Methods: Febrile children 0-5 years of age presenting to the emergency department of a tertiary care pediatric hospital were sampled consecutively. The accuracy of the axillary temperature measured at presentation was evaluated using logistic regression models to generate receiver operating characteristic curves. Reference standard tests for serious bacterial infection were standard microbiologic/radiologic tests and clinical follow-up. Age, clinicians' impression of appearance of the child (well versus unwell) and duration of illness were assessed as possible effect modifiers. Results: Of 15,781 illness episodes 1120 (7.1%) had serious bacterial infection. The area under the receiver operating characteristic curve for temperature was 0.60 [95% confidence intervals (CI): 0.58-0.62]. A threshold of ≥38°C had a sensitivity of 0.67 (95% CI: 0.64-0.70), specificity of 0.45 (95% CI: 0.44-0.46), positive likelihood ratio of 1.2 (95% CI: 1.2-1.3) and negative likelihood ratio of 0.7 (95% CI: 0.7-0.8). Age and illness duration had a small but significant effect on the accuracy of temperature increasing its "rule-in" potential. Conclusion: Measured temperature at presentation to hospital is not an accurate marker of serious bacterial infection in febrile children. Younger age and longer duration of illness increase the rule-in potential of temperature but without substantial overall change in its test accuracy.