← Longevity & Supplement Guides
The Invention of Normal: Where 37 °C Came From
In the early 1990s a physician in Baltimore borrowed a nineteenth-century thermometer, carried it back to his laboratory, and checked it against a calibrated modern one. The instrument had belonged to Carl Reinhold August Wunderlich, and it was in the collection of the Mütter Museum in Philadelphia; the curator offered it for the experiment. It was long — the accounts describe something closer to a foot of glass than to the stub in your bathroom drawer — and it read high.
The number that instrument helped establish is printed on thermometer packaging on two continents. 98.6 °F. Elsewhere, 37 °C. It is one of the few pieces of physiology almost everyone can recite from memory, and the closest thing modern medicine has to a folk constant.
It is also an average of armpit readings, taken in a German teaching hospital in the 1860s, with instruments that ran warm. That, on its own, would be a small and forgivable story about old glass. The larger story is stranger: the man who produced the number never claimed the body sat at it. He published a range, and a daily rhythm, and a list of the ways people differ from one another. We kept the midpoint and threw away the distribution — and then, for the next century and a half, measured ourselves against a figure that was never meant to be a constant.
Before the number, the fever was the disease
It is easy to assume that thermometers arrived in medicine and medicine immediately used them. They did not. Santorio Santorio put a numerical scale on an air thermoscope in the early 1600s and applied it to the human body. Anton de Haen, teaching in Vienna in the mid-1700s, argued for taking temperature at the bedside as a matter of routine. Both were, in the useful sense, correct. Both were largely ignored. Something like two centuries passed between the instrument being available and the instrument being trusted.
Part of the reason is that there was no obvious question for it to answer. Fever, in the medical thinking of the period, was not a sign of something else — it was the thing itself, a family of diseases with names and courses and treatments of their own. You did not need a number to tell you a patient was burning. And a reading that took a quarter of an hour to settle, on an instrument that had to be held in place and read in situ, was not free. It cost the one resource a ward physician never had.
What Wunderlich did was change the question. He was a professor at Tübingen from 1846 and then at Leipzig from 1850, and he belonged to a generation that wanted medicine to look more like physics: measure the thing, plot the thing, and let the pattern make the argument. Temperature, in his hands, stopped being the disease and became a sign — a curve you could put at the foot of a bed and read across days. That reframing, not the number, is his real legacy. The temperature chart is his invention in the way the thermometer is not.
What Wunderlich actually wrote down
His 1868 book, Das Verhalten der Eigenwärme in Krankheiten — translated into English three years later as On the Temperature in Diseases: A Manual of Medical Thermometry — is usually credited with something on the order of a million axillary readings taken from some 25,000 patients. It is an enormous piece of nineteenth-century data collection, done one armpit and fifteen to twenty minutes at a time.
Here is the part that gets left out. Wunderlich identified 37.0 °C as the mean temperature of healthy adults, and in the same work stated that the range around it ran from roughly 36.2 °C to 37.5 °C. He described a daily rhythm, with a trough between about 2 and 8 in the morning and a peak between 4 and 9 in the evening. He noted that women tended to run slightly warmer. He suggested older people ran cooler. He set readings above 38.0 °C as suspicious — the word in the English edition is "probably febrile" — rather than as a bright line between health and illness.
In other words: a mean, a spread, a rhythm, and a set of caveats about who you are. Every one of those four things has held up in some form. Only the first one survived into public memory.
The armpit is not the core
Wunderlich's preferred measurement site was the axilla, and the axilla is a poor stand-in for the inside of a person. It sits at the boundary between what your body is doing and what the room is doing. A modern systematic review of 36 prospective studies — 7,636 healthy adults, more than 9,000 measurement sites — put the mean axillary temperature at 35.97 °C, with rectal at 37.04 °C and oral at 36.57 °C. Read that again in the context of the folk constant: measured today, at the site Wunderlich actually used, healthy adults average about a full degree Celsius below the number he named as the physiological point.
So how did an armpit produce 37.0? At least partly through the glass. This is where the thermometer in the Baltimore laboratory comes back. Tested against a modern digital instrument, Wunderlich's own thermometer read high — by around three degrees Fahrenheit, the physician who ran the test told a reporter years later. Published estimates of the offset in his instruments vary; accounts put it somewhere between roughly 1.4 °C and 2 °C. The exact figure is contested and probably unrecoverable. The direction is not.
None of this makes Wunderlich a fool. He was measuring consistently with the tools he had, in a period when consistency was the achievement. What it makes him is calibrated to his own equipment — which is a description of every measurement anyone has ever taken, including the one on your wrist last night.
How 37.0 became 98.6, and why the extra digit matters
The conversion is where a reasonable average acquired the manners of a law. 37.0 °C, carried into Fahrenheit, is 98.6 °F. Two significant figures went in; three came out. The decimal in 98.6 is not a measurement. It is arithmetic residue, and it is the single most persuasive digit in the history of consumer physiology.
Precision reads as authority. A range invites you to ask where in it you sit; 98.6 does not invite anything. It has the shape of a specification — the temperature a body is supposed to be — and once a number has that shape, it stops being a summary of people and starts being a standard people fail to meet. Anyone who has watched a parent read 99.1 on a drugstore thermometer and go quiet has seen the cost of a spurious decimal place.
Figure · what the average hides
The body has never had one temperature — including in 1868
Three things a single number throws away
The hour. Mean oral temperature moves through the day with a trough around 6 am and a peak between 4 and 6 pm, averaging about 0.5 °C of swing. In the 1992 audit the 99th percentile was 37.2 °C at 6 am and 37.7 °C at 4 pm — the same person, the same health, a different ceiling.
The site. Means across 36 studies: rectal 37.04 °C, oral 36.57 °C, axillary 35.97 °C. A reading without a site is not a number.
The person. In 35,488 outpatients, age, sex, height, weight, ethnicity and comorbidities together explained only 8.2% of the differences between individuals' baseline temperatures. The rest was simply who they were.
Band ends are each source's own published limits and the definitions differ — Wunderlich's stated range, Mackowiak's observed minimum and maximum, Obermeyer's 95% range, Geneva's mean ± 2 SD. They are not interchangeable and are shown together to compare position, not spread. Sources: Wunderlich 1868/1871; Mackowiak, JAMA 1992; Obermeyer, BMJ 2017; Geneva, Open Forum Infectious Diseases 2019.
1992: somebody finally checked
For more than a century, almost nobody re-ran the measurement. Wunderlich's work was so comprehensive, and so early, that it functioned less like a finding than like furniture. The 1992 study that finally audited it came out of the University of Maryland: 148 healthy adults aged 18 to 40, oral temperatures taken one to four times a day for three days, 700 readings in total, on an electronic thermometer with a steady-state error under a tenth of a degree.
The mean was 36.8 °C, not 37.0. The upper limit of normal came out at 37.7 °C, not 38.0. Readings ran from 35.6 °C to 38.2 °C in healthy people going about their day. And 37.0 °C exactly — the folk constant, the specification, the number on the box — accounted for 8% of the 700 observations. Not the centre of gravity. Just one bin among many.
The authors' conclusion was blunt for a journal: 37.0 °C should be abandoned as a concept relevant to clinical thermometry. That was thirty-four years ago. Check the packaging in your bathroom.
We also appear to be cooling
Then it gets stranger. In 2020 a Stanford group assembled three very different bodies of data — 83,900 readings from Union Army veterans of the American Civil War, 15,301 from a 1970s national health survey, and 578,222 from a hospital system between 2007 and 2017 — and looked at temperature by the decade a person was born rather than the decade they were measured.
Mean body temperature declined monotonically, by about 0.03 °C per birth decade, across two centuries. Men born in the early 1800s ran roughly 0.59 °C warmer than men measured in the 2000s. The obvious objection is that this is a story about thermometers, not bodies — and the authors' answer is that the same downward slope appears within the Civil War cohort, measured with a single era's instruments, which is hard to explain as drifting calibration.
Why we might be cooler is unsettled: less chronic infection and inflammation across the population, better-heated rooms, less metabolic work spent holding a line against cold. It is a real open question, and worth holding loosely. But it does something specific to the folk constant. A number derived in 1868 is not merely mis-sited and mis-calibrated. It may also be an accurate description of a population that no longer exists.
Your normal is yours, and it is remarkably stable
The most useful modern paper on this is not about the average at all. In 2017 a team analysed 243,506 temperature readings from 35,488 outpatients — people with no infection diagnosis and no antibiotic prescription, in whom temperature was expected to be unremarkable. The mean was 36.6 °C; 95% of readings fell between 35.7 °C and 37.3 °C.
The finding that matters is the structure underneath. Individual baselines were persistent and distinguishable — some people simply run warm and keep running warm — and everything the researchers could measure about a person accounted for only 8.2% of that variation. Older adults ran slightly cooler, by about 0.021 °C per decade of age. Women averaged marginally warmer than men. All of it small; all of it dwarfed by the unexplained, durable individuality.
Then a detail that should make anyone thoughtful: after adjusting for everything they had, a one-standard-deviation higher individual baseline — 0.149 °C — was associated with 8.4% higher one-year mortality. That is an observational association in a clinic population, not a mechanism and not a reason to take your temperature nervously. But it is a strong hint that your baseline is not noise around a universal set point. It is a signal about you, and the universal number is what has been hiding it.
What the wrist actually measures
Which brings us to the device you may be wearing. Modern wearables that report temperature are almost never measuring core temperature; they are measuring skin at the wrist or finger, which is a peripheral signal shaped by blood flow, sleeves, duvets and room air. That is why decent ones report a deviation from your own baseline rather than an absolute number with a decimal point — a design choice that is, quietly, a correction of the 98.6 error.
There is good evidence that the shape is where the information lives. In 91,462 UK Biobank participants who wore a research-grade wrist device for a week, the amplitude of the daily temperature rhythm — the size of the swing between the low and the high — carried associations with later health outcomes. The authors were explicit about why they used amplitude rather than the average: the mean wrist temperature showed strong device-specific bias, while the amplitude was robust across devices. The absolute number was the contaminated part. The rhythm was the durable part.
Wunderlich, it turns out, was working the right way and remembered for the wrong thing. He took repeated measurements on the same person and read them as a curve. That is what a temperature chart is. The number at its centre was the least interesting output of the method, and it is the only part that made it into common knowledge.
The dogma that outlived the audit
Some of Wunderlich's figures are still load-bearing. A 2020 review of fever in the infectious-disease literature notes that 38.0 °C remains the working threshold in clinical practice — the same number he set as "suspicious" in 1868 — while pointing out that modern baselines are lower, which logically ought to move the line. Wearable temperature features are wellness estimates and nothing more: they cannot diagnose anything, and no reading from a wrist is a clinical measurement.
This is how measurement dogma persists. Not through conspiracy or laziness, but because a number that is embedded in charts, textbooks, packaging, software defaults and the memory of everyone who has ever had a child is more expensive to change than to keep. Numbers stay in use long after the evidence under them has moved, because their real function was never accuracy. It was coordination.
Which is a reason to hold every number we give you loosely, too — the sleep score, the readiness figure, the biological age. We have written about how a single sleep number destroys information and about how much confidence biological-age tests actually earn. 98.6 is the oldest member of that family, and the most beloved. It should be studied as a cautionary tale, not as a constant.
What to do with a number you can't quite trust
Learn your own. This week, take your temperature at the same time of day — morning is easiest — with the same thermometer, at the same site, for five or six days while you feel well. Write down the readings. You are not looking for 37.0. You are building the only reference range that applies to you, and after a few days you will be able to recognise a genuine departure from it without consulting a number set in Leipzig.
Then extend the habit to everything on your wrist. Read the deviation, not the digit; read the trend across a week, not last night's figure; and treat the day-to-day shape — your resting heart rate, your overnight temperature swing, your HRV — as the part carrying signal. The Agen app frames every reading against your own rolling baseline for exactly this reason, and the Agen Band exists to make the baseline cheap enough to have. The point of measurement is to correct fantasy, not to replace how you actually feel.
Two honest limits. Individual baselines shift with the menstrual cycle, illness, alcohol, a warm bedroom and the hour you stand up, so a week of readings is an orientation and not a calibration. And none of this is clinical: 38.0 °C is still the threshold your doctor works to, a reading you are worried about is a conversation for them and not for a wearable, and anyone who feels genuinely unwell should stop reading graphs and make the call.
The bottom line
Carl Wunderlich did something admirable — he turned a curiosity into a clinical instrument and taught medicine to plot a person over time. What went wrong afterwards was not his measurement but our editing. We took a mean, a range, a rhythm and a set of individual caveats, and kept the mean. The conversion into Fahrenheit added a decimal place that was never measured, and the decimal place gave the average the authority of a law.
The correction is available and unglamorous. Your temperature is a distribution with a daily shape and a personal centre, and so is almost every other number you track. Find your own middle, watch the shape, and let the population average be what it always was: a useful summary of other people.
Educational content, not medical advice. This article does not diagnose, treat or prevent any condition. Agen products are wellness tools, not medical devices. If you are unwell, or concerned about a temperature reading, speak to a qualified healthcare professional.
