BMR Equations Compared: Which Formula Is Most Accurate?

Mifflin-St Jeor is the BMR equation with the best published track record for healthy adults: a 2005 systematic review found it predicted resting metabolic rate within 10% of the measured value in more people than any of the other three common equations it compared. But "most accurate" is a statement about groups, not about you. Run three different people through five equations and the answers disagree by 147 to 251 kcal a day:

All values are kcal per day. Person A's five answers span 147 kcal, person B's 163 kcal and person C's 251 kcal, and the equations fall in a different order for each of the three people. The spread is the honest size of the uncertainty before anyone has guessed an activity level.

The five equations, as their authors published them

Each equation below predicts energy use at rest, in kcal per day, from weight in kilograms, height in centimetres and age in years, or from lean or fat-free mass in kilograms. They differ in where the numbers came from.

The Harris-Benedict coefficients are copied from the authors' own 1918 summary in the Proceedings of the National Academy of Sciences, where "h" is total heat production per 24 hours. The full Carnegie Institution monograph followed in 1919, which is why the Mifflin and Cunningham abstracts date the equations 1919. The Mifflin-St Jeor equation is from the 1990 paper's abstract in the American Journal of Clinical Nutrition. The paper gives a single combined form first, 9.99 × weight + 6.25 × height − 4.92 × age + 166 × sex − 161, with sex coded 1 for men and 0 for women, and then the rounded versions in the table, which its authors say did not change the predictive value. The two forms differ by a few kcal: for person B the combined form gives 1,858 against 1,855 for the rounded one.

The two lean-mass rows need more care, because their names travel loosely. J. J. Cunningham published two different equations in the same journal. In 1980 he reanalysed the Harris-Benedict data and proposed BMR = 500 + 22 × lean body mass. In 1991, after a review of later studies of resting energy and body composition, he proposed a general equation, REE = 370 + 21.6 × fat-free mass, which he reported explains 65% to 90% of the variation in resting energy. The equation that calculators and fitness papers call Katch-McArdle is the same 370 + 21.6 × fat-free mass; a 2025 exercise study, for example, prints it under that name. Some later papers print the 1980 equation in a more precise form, 21.6 × fat-free mass + 501.6, so check which version a calculator uses before comparing its output with this table. Both lean-mass columns here are fed the same fat-free mass, weight × (1 − body fat), and the 1980 equation comes out 147 to 157 kcal above the 1991 one for these three people.

The 1984 Roza and Shizgal coefficients are not printed in that paper's abstract. The values in the table are as reproduced in a 2024 peer-reviewed study that used them, cited in the sources.

Three things are worth noticing before any arithmetic:

• Mifflin-St Jeor has one shape for both sexes. The only sex difference is a fixed 166 kcal offset. Harris-Benedict fits separate slopes for men and women, which is why it gives very different weight and age effects: in the 1918 equations each extra kilogram adds 13.75 kcal for a man but 9.56 kcal for a woman.
• The Harris-Benedict constant for women is about ten times the constant for men, 655 against 66. Each equation is a separate straight line fitted to its own sample, so a constant only means something together with the slopes beside it. Comparing constants across these equations tells you little on its own.
• The lean-mass equations ignore age, height and sex. They assume that lean tissue accounts for the differences. Cunningham's 1980 reanalysis is the basis for that assumption: it found lean body mass to be the single predictor and said the presumed influences of sex and age "add little". How well that holds for a given person depends on how good the body-fat figure is, which a later section works through.

Three people, five answers

The dot plot puts the opening table on one scale. Each row is one person. The distance between the lowest and highest dot is the disagreement between equations, before any measurement.

A few patterns hold across all three people, and one does not.

Harris-Benedict lands above Mifflin-St Jeor for all three, by 74 kcal for person A, 114 kcal for person B and 102 kcal for person C on the 1918 version. That fits what Mifflin's team reported when they tested the older equations on their own sample of 498 adults: the Harris-Benedict equations overestimated measured resting energy expenditure by 5% on average. A gap of 5% is about 66 kcal for person A and 93 kcal for person B. The revised 1984 coefficients move by no more than 10 kcal for any of the three people here, so the revision does not close the gap.

The two lean-mass equations sit apart from each other. For person A, with 28% body fat, Katch-McArdle gives the lowest number and Cunningham 1980 the highest, and they bracket the weight-based equations. For person C, a 70-year-old woman at 40% body fat, Cunningham 1980 gives 1,490 kcal, 251 kcal above Mifflin-St Jeor, and Katch-McArdle gives 1,342 kcal, 103 kcal above it. Her 45 kg of fat-free mass is about the same as person A's 43.2 kg, so both lean-mass equations treat the two women as almost identical, despite a 40-year age gap and 15 kg more body weight for person C. Mifflin-St Jeor reduces the estimate by 5 kcal for every year of age and adds 10 kcal for every kilogram, and it puts person C 81 kcal below person A.

Which equation is "right" for these three people cannot be read off the chart. They are illustrations with stated inputs, not measured subjects. The 2005 review names the method that would settle it: estimation errors "would be eliminated by valid measurement of RMR with indirect calorimetry". Each equation on this page was derived from measurements on a group of people and is then applied to someone who was not measured.

What each equation was built on

An equation reflects the people it was fitted to. The samples behind these equations are small, old or narrow by today's standards.

The Harris-Benedict subjects were measured more than a century ago, and a 2026 study that reproduces the equation lists their mean age as 29. The original paper says plainly what the equations were for: standard constants "until those based on more extensive series of data are available." Mifflin-St Jeor's sample was about twice the size, ran to age 78 and included 234 people with obesity.

The 2005 review that is usually quoted behind the "Mifflin is most accurate" claim also had a warning. Its authors, writing in the Journal of the American Dietetic Association, found that older adults and US-residing ethnic minorities were underrepresented both in developing the equations and in testing them. For members of groups that are greatly underrepresented, they wrote that "a high level of suspicion regarding the accuracy of the equations is warranted." For an individual reader, that sentence matters more than the ranking.

The Mifflin paper also tested a fat-free-mass version on its own sample: REE = 19.7 × fat-free mass + 413. That is a third lean-mass equation, with a smaller slope than the 21.6 and 22 of Cunningham's two. For person B's 66.5 kg of fat-free mass it gives 1,723 kcal, 83 kcal below Katch-McArdle. Equations that look alike were fitted to different groups and do not agree in detail.

How far off a prediction can be

"Within 10%" is the yardstick the 2005 review and the 2026 study below both use: a prediction counts as accurate if it is between 90% and 110% of the measured value. For person B, that window runs from 1,670 to 2,041 kcal around the Mifflin-St Jeor figure of 1,855. That is a 371 kcal range, wider than the whole spread of the five equations for that person.

Three sources put a size on the error.

The 2005 systematic review (Frankenfield, Roth-Yousey and Compher, Journal of the American Dietetic Association) looked at the four equations it identified as most commonly used in clinical practice: Harris-Benedict, Mifflin-St Jeor, Owen and the WHO/FAO/UNU equation. Mifflin-St Jeor predicted resting metabolic rate within 10% of measured in more nonobese and obese people than any other, and it had the narrowest error range. The review found no validation work concentrating on individual errors for the WHO/FAO/UNU equation. The per-equation percentages sit in the full paper's tables, which we did not read, so this page uses only the ranking stated in the abstract.

Roza and Shizgal put the precision of Harris-Benedict for a normally nourished person at plus or minus 14%. For person B's Harris-Benedict figure of 1,969 kcal, that is about 276 kcal in either direction.

A 2026 study of 189 community-dwelling older adults in Shanghai (Cai and colleagues, published in Nutrients) shows what happens when a population differs from the original samples. In its validation group of 57 people, Mifflin-St Jeor landed within 10% of measured resting rate for 49.1% of them, Harris-Benedict for 26.3% and Cunningham for 10.5%. The Cunningham version the study tested is the 21.6 × fat-free mass + 501.6 form. All three overestimated on average, by 113 kcal, 177 kcal and 298 kcal respectively. In women, Mifflin-St Jeor showed no significant average bias, while in men all 11 published equations the study tested overestimated. The authors tie the error to lower fat-free mass and body weight in their participants than in the Western samples the equations came from.

Error does not always run the same way. A 2024 paper on resting rate in people with overweight and obesity (Gitsi and colleagues, Journal of Clinical Medicine) describes a 30-year-old woman with a BMI of 59.6 whose measured resting rate was 2,726 kcal. Harris-Benedict predicted 2,274 kcal, 16.6% too low, and Katch-McArdle 1,752 kcal, 35.7% too low. The authors attribute the shortfall to the equations leaving fat mass out. So an equation can overshoot a small older adult by hundreds of kcal and undershoot a person with severe obesity by more.

Taken together: Mifflin-St Jeor is the equation the 2005 review found most likely to land within 10% for nonobese and obese adults, and even so a 10% error is ordinary rather than a sign that something went wrong. For older adults and ethnic minorities, the groups the review found underrepresented, and at the extremes of body fat shown above, the published accuracy figures may not carry over. These equations estimate resting energy use for healthy adults. They are not a diagnosis, and the review's own conclusion points clinicians who need a precise figure to indirect calorimetry rather than a formula.

From resting rate to daily needs: multipliers against the 2023 equations

A BMR or resting rate is only part of the day. Total energy expenditure (TEE, or TDEE in calculator language) adds the energy of digestion and all movement. The TDEE calculator on this site gets there with the multiplier method: it multiplies the Mifflin-St Jeor figure by an activity factor of 1.2 for sedentary, 1.375 for light exercise, 1.55 for moderate, 1.725 for heavy exercise and 1.9 for an athlete.

The National Academies took a different route in its 2023 Dietary Reference Intakes for Energy. Its committee built equations that predict total energy expenditure directly, from a combined database of 8,600 doubly labeled water observations across all ages (5,456 of them adults 19 and over). The report describes doubly labeled water as considered "the benchmark standard for measurement of total energy expenditure (TEE) in humans under free-living conditions." These are not BMR equations, and their output should not be compared with a BMR figure. For adults 19 and over, with height in centimetres and weight in kilograms:

The report gives their accuracy: for men a root mean square error of 339 kcal per day and a mean absolute percentage error of 9.4%; for women 246 kcal and 8.7%. In other words, even an equation fitted to thousands of direct measurements of total energy use misses by around 9% on average.

Putting the two approaches side by side shows the main practical problem with the multipliers:

For person A, the 2023 equation for the inactive category gives 2,021 kcal. Mifflin-St Jeor times the sedentary 1.2 factor gives 1,584 kcal, 437 kcal lower. For person B the same gap is 549 kcal and for person C 401 kcal. Dividing each 2023 estimate by the Mifflin-St Jeor resting figure shows why: the National Academies' "inactive" category works out to about 1.5 times the resting rate for all three people (1.53, 1.50 and 1.52), and its "low active" category to 1.62 to 1.66. The report defines PAL as the ratio of total energy expenditure to BMR, and its percentile table for adults aged 19 to 70.99 puts the median PAL at 1.68 and the 10th percentile at 1.39.

So a factor of 1.2 applied to a resting estimate sits well below the PAL of 1.39 that the report found at the 10th percentile for adults under 71. The two methods are closer at the top end: "very active" in the 2023 equations and the 1.9 multiplier agree within 100 kcal for all three people (44, 95 and 60 kcal apart).

This does not make the 2023 equations right for any individual. Choosing a PAL category is still a judgement, and moving one category up or down changes the answer by 136 to 417 kcal for these three people, about as much as or more than the choice between BMR equations. The calorie calculator is the tool on this site for running a daily figure on your own inputs; the table above shows how much that figure depends on the activity choice.

When lean-mass equations help and when they hurt

Cunningham and Katch-McArdle replace weight, height, age and sex with a single number: fat-free mass. The case for them is in the data. Cunningham's 1980 reanalysis found lean body mass to be the single predictor of BMR, and in Mifflin's sample fat-free mass alone explained 64% of the variation in measured resting energy, against 56% for weight alone. The catch is the input. A calculator computes fat-free mass as weight × (1 − body fat percentage), so any error in the body fat reading passes straight through to the answer.

The arithmetic of that error is simple. Every kilogram of fat-free mass is worth 21.6 kcal in Katch-McArdle and 22 kcal in Cunningham 1980. A body fat reading that is off by 5 percentage points shifts fat-free mass by 5% of body weight. For our three people, using Katch-McArdle:

Five points is used here as an illustration of sensitivity, not as the error of any particular device. At that size, the uncertainty from the body-fat reading alone for person B is larger than the whole 163 kcal spread among the five equations. A lean-mass equation can only be as good as the body fat figure it is given, and the swing grows with body weight.

The documented failures in the studies above run in both directions. At very high body fat, Katch-McArdle missed the BMI 59.6 case above by 35.7%, which the authors put down to leaving fat mass out. In the 2026 Shanghai study of older adults, the Cunningham form tested overestimated measured resting rate by 298 kcal on average. The lean body mass calculator and the body fat calculator give the inputs, and the BMR calculator runs Mifflin-St Jeor. Comparing its answer with Katch-McArdle on your own fat-free mass is the quickest way to see whether the body fat figure is driving the difference.

Checking an estimate against a scale

No equation on this page can be validated for one person without measurement. Body weight over time is something anyone can record, and energy balance links it to the estimate. The link is slower than the familiar rule suggests.

The familiar rule is 3,500 kcal per pound: eat 500 kcal a day less and lose a pound a week, since 500 × 7 is 3,500. A 2011 analysis in The Lancet by Kevin Hall and colleagues at the US National Institutes of Health called that static rule erroneous, because it ignores how energy expenditure adapts as weight changes. Their model gives a different rule of thumb for an average overweight adult: every change of energy intake of 100 kJ per day leads to an eventual weight change of about 1 kg, equivalently 10 kcal per day per pound, with half of the change reached in about a year and 95% in about three years.

That rule turns equation error into something you can picture. Suppose an estimate of daily energy use is 10% too high for person B: about 186 kcal at his Mifflin-St Jeor resting figure, or about 278 kcal on the 2,775 kcal that the 2023 inactive equation gives him. Under the Hall rule, eating to a figure that is 186 kcal too high would eventually move weight by about 19 pounds, and 278 kcal by about 28 pounds. Treating the approach as a curve that covers half the distance each year (a simple half-life approximation, not Hall's own projection), the first month would show only about 1.0 to 1.6 pounds of it and the first three months about 3.0 to 4.4 pounds.

Three practical consequences follow for anyone checking an estimate:

1. Months, not days. On those numbers, a 186 to 278 kcal error builds up to only 1 to 1.6 pounds in a month. A trend over two to three months, taken at the same time of day and averaged, carries far more of the signal than a week does.
2. Body fat changes the timescale. The Hall analysis found that adults with more body fat have a larger expected weight change for the same change in intake, and take longer to reach their new steady weight. The same 186 kcal error means different things for persons A, B and C.
3. Use a model, not a constant. The 2011 paper presented a web-based simulator built on the same dynamic model. The NIH's National Institute of Diabetes and Digestive and Kidney Diseases publishes a free Body Weight Planner that lets adults 18 and over make calorie and physical activity plans to reach a goal weight within a set time and maintain it afterwards. It is a better reference than any fixed number of calories per pound.

This is measurement, not a plan, and it is not medical advice: none of the arithmetic here says what anyone should eat. Health conditions, medication, pregnancy and growth all change energy needs in ways these adult resting equations were not built for, and those are questions for a clinician or registered dietitian. What the arithmetic does show is that the gap between Harris-Benedict and Mifflin-St Jeor, 74 to 114 kcal for the three people here, is smaller than the 136 to 417 kcal that one step of activity category moves the 2023 estimate, and a few months of careful weighing can reveal errors of that size.

For your own Mifflin-St Jeor figure, start with the BMR calculator, then set the other equations from the table at the top of this page next to it. Where they disagree widely, the activity estimate and the scale will usually tell you more than the choice of formula. The fitness hub collects the heart-rate, pace and body-composition tools that sit around it.

Sources

• Harris, J. A., and F. G. Benedict, "A Biometric Study of Human Basal Metabolism," Proceedings of the National Academy of Sciences 4(12):370-373, 1918 (equations, 136 men and 103 women, 12 to 14 hours after the last meal, table ranges), via Internet Archive scan of the JSTOR early journal content: https://archive.org/details/jstor-83837 (checked October 28, 2026)
• Mifflin, M. D., S. T. St Jeor, et al., "A new predictive equation for resting energy expenditure in healthy individuals," American Journal of Clinical Nutrition 51(2):241-247, 1990, PubMed abstract (498 subjects, equations, R², Harris-Benedict overestimate of 5%, fat-free mass and weight equations): https://pubmed.ncbi.nlm.nih.gov/2305711/ (checked October 28, 2026)
• Roza, A. M., and H. M. Shizgal, "The Harris Benedict equation reevaluated," American Journal of Clinical Nutrition 40(1):168-182, 1984, PubMed abstract (98 added subjects, precision of plus or minus 14%): https://pubmed.ncbi.nlm.nih.gov/6741850/ (checked October 28, 2026)
• Cunningham, J. J., "A reanalysis of the factors influencing basal metabolic rate in normal adults," American Journal of Clinical Nutrition 33(11):2372-2374, 1980, PubMed abstract (223 subjects, BMR = 500 + 22 × LBM): https://pubmed.ncbi.nlm.nih.gov/7435418/ (checked October 28, 2026)
• Cunningham, J. J., "Body composition as a determinant of energy expenditure: a synthetic review and a proposed general prediction equation," American Journal of Clinical Nutrition 54(6):963-969, 1991, PubMed abstract (REE = 370 + 21.6 × FFM, 65% to 90% of variation): https://pubmed.ncbi.nlm.nih.gov/1957828/ (checked October 28, 2026)
• Frankenfield, D., L. Roth-Yousey, and C. Compher, "Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review," Journal of the American Dietetic Association 105(5):775-789, 2005, PubMed abstract: https://pubmed.ncbi.nlm.nih.gov/15883556/ (checked October 28, 2026)
• National Academies of Sciences, Engineering, and Medicine, Dietary Reference Intakes for Energy, 2023, chapter 5, Table 5-3 (PAL percentiles), Table 5-4 (PAL categories) and Table 5-5 (TEE prediction equations, error statistics): https://nap.nationalacademies.org/read/26818/chapter/7 (checked October 28, 2026)
• Hall, K. D., et al., "Quantification of the effect of energy imbalance on bodyweight," The Lancet 378(9793):826-837, 2011 (rule of thumb, half time of about one year, critique of the 3,500 kcal rule, web-based simulator): https://pubmed.ncbi.nlm.nih.gov/21872751/ (full text PMC3880593, checked October 28, 2026)
• Cai, Z., et al., "Prediction Equations to Estimate Resting Metabolic Rate in Healthy, Community-Dwelling Chinese Older Adults," Nutrients 18(2):344, 2026 (Harris-Benedict sample age, Cunningham form tested, validation results): https://doi.org/10.3390/nu18020344 (full text PMC12844918, checked October 28, 2026)
• An, S., W. Eo, et al., "Basal metabolic rate by FAO/WHO/UNU as a prognostic factor for survival in patients with gastric cancer: A cohort study," Medicine, 2024 (Roza and Shizgal coefficients as reproduced): https://doi.org/10.1097/MD.0000000000040665 (full text PMC11596422, checked October 28, 2026)
• Grivas, G. V., et al., "Comparative analysis of combined exercise programs in middle-aged obese males: impact on energy expenditure, body composition, and metabolic rate," Frontiers in Sports and Active Living, 2025 (equation used under the Katch-McArdle name): https://doi.org/10.3389/fspor.2025.1533030 (full text PMC11996765, checked October 28, 2026)
• Gitsi, E., A. Kokkinos, et al., "The Relationship between Resting Metabolic Rate and Body Composition in People Living with Overweight and Obesity," Journal of Clinical Medicine 13(19):5862, 2024 (BMI 59.6 case): https://doi.org/10.3390/jcm13195862 (full text PMC11477793, checked October 28, 2026)
• National Institute of Diabetes and Digestive and Kidney Diseases, Body Weight Planner: https://www.niddk.nih.gov/bwp (checked October 28, 2026)
• The three people are illustrations with stated inputs. Every equation output, the multiplier comparison, the body-fat swing and the weight projections were calculated by CalculatorVillage from the published equations.