Mifflin-St Jeor wins the general case and loses one population badly#
For most adults, use Mifflin-St Jeor. The systematic review that set the four clinically common equations side by side — Harris-Benedict, Mifflin-St Jeor, Owen, and WHO/FAO/UNU — ranked Mifflin first on the one criterion that matters to an individual — the share of people whose measured resting rate it landed inside a 10 percent band of, in nonobese and obese samples alike — and gave it the tightest error range of the four4. Measured head-to-head in 125 healthy women aged 20 to 57 with BMIs spanning 17 to 44, Mifflin landed inside that 10 percent band for 71 percent of them against Harris-Benedict's 61 percent, and carried no group-level bias at all: 0 kcal/day (SD 153) against Harris-Benedict's +78 kcal/day (SD 157)5.
Then the ranking turns over. In 85 adults aged 65 and older living with obesity — 144 separate measurements against a metabolic cart — Harris-Benedict was within 10 percent for 53.5 percent of measurements and Mifflin for only 43.1 percent, with Mifflin falling below the measured value 44.4 percent of the time6. Neither actually won there: the WHO equation did, at 59.0 percent, which still means it missed two measurements in five. So the useful version of "which equation is best" has a population attached to it. What follows is where the two formulas came from, what the head-to-heads measured, and what separates the studies that disagree. How to calculate your TDEE has the arithmetic you wrap around whichever one you choose.
Two equations, seventy-two years apart#
Harris and Benedict published A Biometric Study of Human Basal Metabolism in the Proceedings of the National Academy of Sciences in 1918, out of the Carnegie Institution's Nutrition Laboratory1; the equations are conventionally dated 1919 and still carry that year in the clinical literature. They predict resting expenditure from weight, height, age and sex — exactly what every consumer calculator asks you for a century later.
The formula was audited properly once, in 1984. Roza and Shizgal went back to the indirect-calorimetry data behind the original equations — 239 subjects, plus a further 98 from Benedict's later published work covering a wider age range — recomputed the coefficients, and reported that the Harris-Benedict equations estimate resting energy expenditure in a normally nourished subject with a precision of 14 percent2. Fourteen percent on a 1,500-calorie resting rate is a band about 210 calories wide — our arithmetic on their figure — and that was the verdict of the people trying to rescue the equation, not of its critics.
Mifflin-St Jeor arrived in 1990 out of a straightforward exercise: measure a lot of modern people and fit a new line. Resting energy expenditure was measured by indirect calorimetry in 498 healthy adults — 247 women, 251 men, aged 19 to 78, mean age 45 ± 14, of whom 264 were normal-weight and 234 had obesity — and the regression on weight, height, age and sex returned R² = 0.713. The same paper ran the old formula against the new measurements and found that the Harris-Benedict equations overestimated measured resting expenditure by 5 percent (P < 0.01). That 5 percent is most of the practical difference between the two, and it points in a consistent direction: the older equation reads high.
| Harris-Benedict | Mifflin-St Jeor | |
|---|---|---|
| First published | 1918, PNAS | 1990, Am J Clin Nutr |
| Inputs | Weight, height, age, sex | Weight, height, age, sex |
| Development sample | 239 subjects, plus 98 more in the 1984 re-analysis | 498 adults, 19–78 y, 234 with obesity |
| Reported precision | ±14% in normally nourished subjects (Roza 1984) | R² = 0.71 (Mifflin 1990) |
| Known systematic error | Overestimates measured REE by ~5% (Mifflin 1990) | No group-level bias in women (Thom 2020); underpredicts in older adults with obesity (Griffith 2022) |
The head-to-head, population by population#
Set the validations against each other and the pattern is not "Mifflin is better." It is "Mifflin is better in people who resemble the ones it was built from."
| Study | Population | Method | Mifflin within ±10% | Harris-Benedict within ±10% |
|---|---|---|---|---|
| Thom 2020 | 125 women, 20–57 y, BMI 17–44 | Ventilated-hood calorimetry | 71% | 61% |
| Griffith 2022 | 85 adults ≥65 y with obesity | Indirect calorimetry, 144 measurements | 43.1% | 53.5% |
| Frankenfield 2005 | Healthy nonobese and obese adults | Systematic review of four equations | Best of the four | Beaten by Mifflin |
Two details inside Thom's data are worth more than the headline. Mifflin's accuracy actually rose with body mass in that sample — 80 percent of the women with obesity were predicted within 10 percent, its strongest subgroup — while its weakest performance came in the underweight women. And its bias was zero. An equation with no systematic offset is a different animal from one that is merely close: Harris-Benedict's +78 kcal/day means it hands nearly everybody a slightly generous number, in the same direction, every time.
The equation that wins the general case underpredicted resting metabolism in 44 percent of adults over 65 with obesity. "Best equation" is not a property of a formula. It is a property of a formula and a person.
What separates the two results#
Nothing here is contradictory, and staging it as a controversy would misread it. Thom measured women aged 20 to 57. Griffith measured adults averaging 73 years with a mean BMI near 36. Different populations, different answers — the expected result when a regression is asked to work outside the range that fitted it.
What is worth naming is the direction of the failure. Both equations subtract a flat amount for every year of age; Mifflin's coefficient is 4.92 kcal per year, rounded to 5 in the simplified version everyone actually uses. A fixed annual subtraction applied out to age 73 keeps removing calories long after the sample supporting it has thinned, and Griffith's specific failure mode was Mifflin coming in low — in 44.4 percent of measurements. The review that crowned Mifflin flagged this territory in advance, noting that the samples behind these formulas, and the samples used to check them, both thin out badly at older ages4. Griffith's numbers are what that caveat looks like once somebody goes and measures. Those authors' own conclusion was blunter still: in this population, equations may not be reasonable substitutes for calorimetry where accuracy is required.
The lean-mass argument, made by Harris-Benedict's own revisers#
There is a third family of equations — the ones that predict from fat-free mass rather than body weight — and the strongest published case for them comes from the 1984 Harris-Benedict re-analysis itself. Having re-run the original data, Roza and Shizgal concluded that resting energy expenditure is directly related to the size of the body cell mass and is independent of age and sex2.
Read that against the two equations everybody actually uses, both of which carry an age term and a sex term. Age and sex are in there as proxies — cheap stand-ins for how much metabolically active tissue a person is likely to be carrying, because weight, height, age and sex are the four things you can collect without a scanner. That is the case for a lean-mass equation if you are lean and trained. It is also why it is not a free upgrade: you have swapped an assumption about your body composition for a measurement of it, and body-composition measurements carry error of their own. Trading a guess for a poor estimate wins nothing.
Pick one, then stop shopping#
- General adult under 65: Mifflin-St Jeor. Best-supported default, and the only one of the pair with a validated near-zero bias in a healthy adult sample.
- Over 65, or carrying substantial obesity: treat every equation as an opening bid. In the population Griffith measured, even the best-performing formula failed 41 percent of the time.
- Lean, trained, with a real body-composition measurement: a fat-free-mass equation is defensible, on Roza and Shizgal's own logic.
Then notice how small this choice is next to the one that follows it. Whichever resting figure you land on gets multiplied by an activity factor you assigned to yourself off a dropdown, and the gap between two adjacent categories is wider than the gap between these two equations — the case for that is in activity multipliers explained. The equation debate is real but bounded; the multiplier is where the wide error lives, which is the shape of the whole estimate laid out in TDEE explained.
The label on the box will be wrong either way: both formulas were fitted to resting metabolic rate and are sold to you as BMR calculators, a distinction unpacked in BMR vs RMR. What finally settles the number is not the formula at all. It is two to four weeks of your own intake against your own weight trend — the method in finding your maintenance calories.
FAQ#
Should I use Mifflin-St Jeor or Harris-Benedict?#
Mifflin-St Jeor, unless you are over 65 or carrying substantial obesity. In 125 women aged 20 to 57 it was accurate to within 10 percent for 71 percent of participants against Harris-Benedict's 61 percent, and unlike Harris-Benedict it showed no systematic offset. In 85 adults aged 65 and over with obesity that ordering reversed — Harris-Benedict 53.5 percent, Mifflin 43.1 percent — so in older age the defensible answer is that no equation is dependable enough to treat as a fact.
Why does the Harris-Benedict equation usually give a higher number?#
Because it reads systematically high in modern samples. When the Mifflin-St Jeor team measured 498 adults by indirect calorimetry in 1990, the 1919 equations overestimated measured resting expenditure by 5 percent (P < 0.01), and a 2020 validation in 125 women put the overprediction at 78 kcal/day. That is a consistent offset rather than random scatter, which is exactly why it matters: a formula generous to nearly everyone quietly inflates every target built on top of it.
Is a century-old equation still safe to use?#
Usable, and its own revisers put a number on how usable. Re-examining the original calorimetry data in 1984, Roza and Shizgal reported that the Harris-Benedict equations predict resting energy expenditure in a normally nourished person with a precision of 14 percent, and that they become unreliable in malnourished patients. Age is not really the objection. Having been fitted to a different population is.
Sources#
- Harris JA, Benedict FG. A Biometric Study of Human Basal Metabolism. Proc Natl Acad Sci U S A. 1918;4(12):370-373.
- Roza AM, Shizgal HM. The Harris Benedict equation reevaluated: resting energy requirements and the body cell mass. Am J Clin Nutr. 1984;40(1):168-182.
- Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr. 1990;51(2):241-247.
- Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc. 2005;105(5):775-789.
- Thom G, Gerasimidis K, Rizou E, et al. Validity of predictive equations to estimate RMR in females with varying BMI. J Nutr Sci. 2020;9:e17.
- Griffith R, Shean R, Petersen CL, et al. Validation of resting energy expenditure equations in older adults with obesity. J Nutr Gerontol Geriatr. 2022;41(2):126-139.
