OBJECTIVE To assess the ability of parents to subjectively evaluate their children for fever and to compare their assessments with temperature measurements made with the use of a noncontact tympanic (NCT) or rectal glass mercury thermometer. A secondary goal was to assess how well a recently developed definition of fever for NCT thermometers, when used in the ear-equivalent mode (temperature of 37.7 degrees C or more), performed in a clinical situation. METHODS This 6-month prospective observational study employed a convenience sample of 180 children, aged birth to 4 years, who presented to the emergency department of a tertiary care children's hospital. Parents were asked to subjectively assess whether their child had a fever. The child's temperature was then measured with an NCT thermometer (three times in the rectal-equivalent mode and three times in the actual-ear mode). Both the subjective assessment and the NCT temperatures were compared with the rectal temperature measured by a rectal glass mercury thermometer. RESULTS The mean age of participants was 14.6 +/- 11.8 months (range, 2 days to 48 months); 56% were boys. The sensitivity of parental detection of fever by subjective means was 81.8% and the specificity 76.5%. The parent and the rectal glass thermometer agreed 79% of the time (95% confidence interval [CI], 73% to 85%). The sensitivity of the first temperature reading obtained with the NCT thermometer in rectal-equivalent mode was 74.7%, and the specificity was 96.3%. The NCT thermometer and the rectal glass thermometer agreed 84% of the time (95% CI, 78% to 89%). Use of the proposed definition of fever for NCT thermometers, when used in the ear-equivalent mode, caused sensitivity of a single measurement for fever to drop to 53.5%. CONCLUSIONS Parental subjective assessment of fever agreed with the presence of fever as measured by rectal glass thermometer in 79% of cases. Specificity was improved with the use of the NCT thermometer. The recently proposed definition for fever for NCT thermometers, when they are used in the ear-equivalent mode, does not appear to be validated by the current data.