Normal human body temperature is exactly 98.6°F (37°C).

In 1851, a German physician named Carl Reinhold August Wunderlich arrived at the University of Leipzig as professor of medicine and began the most comprehensive study of human body temperature anyone had undertaken. Over the following decade and a half, he recorded temperatures taken from the axilla, the armpit, of roughly 25,000 patients, accumulating more than a million individual measurements. His methodology was rigorous by the standards of the time. His conclusions, published in 1868 in a book titled 'On the Temperature in Disease,' included a figure that would persist in medical textbooks for more than a century: the normal human body temperature is 98.6 degrees Fahrenheit, or 37 degrees Celsius.

The figure was taken as definitive almost immediately. Wunderlich's sample was enormous by nineteenth-century standards; nothing like it had been attempted before. The number 37°C had the comfortable appeal of a round figure in the metric system, suggesting precision and order. Medical education codified it. Textbooks printed it. Parents buying thermometers read it on the packaging. For generations of students from primary school through medical school, 98.6 was not a statistical average with a distribution; it was the single correct answer, the standard against which all deviation was measured as fever or hypothermia.

The problems with Wunderlich's number were not examined carefully until 1992, when Philip Mackowiak and colleagues at the University of Maryland published a study in the Journal of the American Medical Association. Mackowiak's team measured oral temperatures in 148 healthy adults over four days under controlled conditions, using precise electronic thermometers calibrated against a mercury standard. The mean temperature was 98.2°F, lower than Wunderlich's figure. More significantly, fewer than 1 percent of temperatures in the study exceeded 99.9°F, suggesting that the common fever threshold of 100.4°F was also too high. Mackowiak's paper noted that Wunderlich's original thermometers were likely uncalibrated and that his axillary measurement site reads consistently higher than oral or rectal sites, both of which would have produced inflated readings of the true core temperature.

The revisionist momentum gathered in subsequent years. A 2020 study led by Julie Parsonnet at Stanford University, published in eLife, analyzed three large historical datasets spanning from Wunderlich's era to the present and found a consistent downward trend in body temperature over 150 years. The average American adult today, the study estimated, runs approximately 1.06°F lower than the Americans of Wunderlich's cohort, a reduction attributable, Parsonnet and colleagues argued, to reduced rates of chronic infection and inflammation, improved nutrition, and changes in the thermal environments in which people live and work.

The finding carried a consequence for how 98.6 should be understood: not only had the number been inaccurately measured, but the underlying physiological reality it was meant to capture had shifted over time. Even if Wunderlich's measurements had been precise, they would reflect the normal temperature of mid-nineteenth-century German patients, not modern humans in different bodies and environments.

Body temperature also varies significantly by time of day, lower in the early morning, higher in the late afternoon, by measurement site, by age, and by individual. A person who normally runs at 97.4°F will have a significant physiological fever at 99.4°F, while someone who normally runs at 98.8°F will not. The single canonical number collapsed under exactly the kind of individual variation that a truly large dataset reveals.

The 98.6 figure has proved difficult to dislodge from public consciousness. It appears on thermometer packaging, in health education, and in popular media. The number is precise-sounding in a way that the honest answer, body temperature varies by individual, time of day, and measurement site, and the population average is closer to 97.5°F, is not. Precision, even false precision, is easier to remember than a range.