BACKGROUND: A clinically useful temperature measurement method should correlate well with the body's core temperature. Although previous investigators have studied temperature readings from different sites in hypothermic and normothermic patients, none have compared methods specifically in febrile patients. OBJECTIVE: To compare temperature measurement methods in febrile intensive care patients. METHODS: Temperature readings were obtained in rapid sequence from an electronic thermometer for oral and axillary temperature, rectal probe, infrared ear thermometer on "core" setting, and pulmonary artery catheter, approximately every hour during the day and every 4 hours at night. The sample consisted of 13 patients with pulmonary artery catheters and with temperatures of at least 37.8 degrees C. RESULTS: Rectal temperature correlated most closely with pulmonary artery temperature. Rectal temperature showed closest agreement with pulmonary artery temperature, followed by oral, ear-based, and axillary temperatures. Rectal and ear-based temperatures were most sensitive in detecting temperatures greater than 38.3 degrees C. Likelihood ratios for detecting hyperthermia were 5.32 for oral, 2.46 for rectal, and 1.97 for ear-based temperature. Rectal and ear-based temperatures had the lowest negative likelihood ratios, indicating the least chance of a false negative reading. Axillary temperature had a negative likelihood ratio of 0.86. CONCLUSIONS: Rectal temperature measurement correlates most closely with core temperature. If the rectal site is contraindicated, oral or ear-based temperatures are acceptable. Axillary temperature does not correlate well with pulmonary artery temperature. These results underscore the importance of consistency in method when establishing temperature trends, and of awareness of method when interpreting clinical data.