Body Fat Percentage Calculator: the US Navy Circumference Equations and the Durnin-Womersley Skinfold Equations, Each With Its Published Error

Two field methods for estimating the proportion of the body that is fat — the US Navy circumference equations in the centimetre form their authors published, and the Durnin and Womersley four-site skinfold equations — computed side by side so that the gap between them is visible. The published errors are on the page rather than behind it, because they are the point: the Navy equations carry a standard error of estimate of 3.52 percentage points in men and 3.72 in women against underwater weighing, and against DXA in independent samples they have been reported biased by anything from 2.3 to 6.0 points. A single confident figure is the misleading output. There is no category, no range and no target here: no reference distribution that could be reproduced on a commercial site was found, and the page says so instead of inventing one.

Last updated 5 Oct 2026

These are measurements of body size. They are not a judgement about you and they are not a diagnosis. They cannot see muscle, bone, where fat sits, or how healthy you are — two people with the same figure can be in very different health. Treat anything here as one rough signal among many, and speak to a doctor or dietitian before making a substantial change to how you eat.

Two published field estimates of body fat, and the error each one carries

circumferences or four skinfolds, with height, weight, age and the equation set -> an estimated body fat percentage, the other method's figure beside it, and the published interval around both

Which method the headline usesBoth methods are computed from whatever is in the fields below, whichever one the headline shows, so the rows underneath always carry the other method’s figure and the difference between them. That difference is the most useful number on the page: the two methods measure different things about the body — girths at three places against the thickness of subcutaneous fat at four — and they are fitted against the same reference by different regressions, so where they disagree sharply the honest reading is that neither is pinning the answer down.

Which equation setBoth methods were fitted separately on male and female samples and the two sets are different equations, not one equation with a correction term. For the Navy method the measurement sites themselves differ: the men’s equation uses the waist taken AT THE UMBILICUS and the women’s uses the waist at its narrowest point, plus the hip. For the skinfold method the four sites are the same for both and only the regression constants change. This control selects which published equation pair is applied; it is a property of the equations rather than a statement about the person using them.

Age (years)Used only by the skinfold method, which bands age at under 20, 20 to 29, 30 to 39, 40 to 49 and 50 and over, and uses different regression constants in each band. That banding has a consequence the page quantifies in the rows below: the estimate steps on a birthday. With an unchanged set of four skinfolds totalling 50 mm, the men’s equations give 18.9% at 29 and 21.5% at 30, 21.5% at 39 and 24.6% at 40, and 24.6% at 49 and 26.4% at 50 — about 7.5 percentage points of change from age alone between 29 and 50. That is an artefact of banding rather than anything happening in a body on a particular night. The Navy equations take no account of age at all, which is a different kind of error, not a smaller one. The page refuses an age outside 18 to 80: the skinfold constants have a published band for 16 to 19 but this page is for adults, and above 80 neither method has a sample behind it.

Height (cm)Used by the Navy equations, where it enters as a logarithm and acts as the scale against which the girths are read — the same waist on a taller frame implies less fat. Measure it without shoes against a wall at the end of a normal breath. This page is metric throughout: centimetres for every length and millimetres for the skinfolds. The Navy equations were published in centimetres and that is the form implemented here; the inch-coefficient version used in the US Department of Defense instruction is computed in the rows below so you can see how far apart the two forms are.

Weight (kg)Not used by either equation. It is here so that the percentage can be turned into a mass of fat and a fat-free mass in kilograms, which is the form in which the uncertainty becomes legible: a 95% interval of plus or minus seven percentage points on a 75 kg body is about ten and a half kilograms of fat, which is larger than most of the differences anyone is trying to detect. It also lets the page print the BMI for the same body, so the two answers can be read against each other.

Neck circumference (cm)Navy method only. The published protocol is to take it just below the larynx with the tape sloping very slightly downward at the front. It enters the equations as a subtraction from the waist, so it is acting as a crude correction for frame size, and an error here moves the answer in the opposite direction to the same error in the waist. Because it is subtracted rather than averaged, the neck is the measurement whose error the Navy equations tolerate least: at a waist of 90 cm and a neck of 38 cm, one centimetre on the neck is worth about 0.7 percentage points of the estimate for a man, and the page prints that sensitivity for your own figures in the rows below.

Waist circumference (cm)Navy method only, and the two published equations use two different sites, which almost every online implementation of them ignores. The men’s equation was fitted on the circumference at the level of the umbilicus. The women’s was fitted on the circumference at the level of minimal abdominal width, roughly midway between the lower end of the sternum and the umbilicus — the natural waist, which on most people is a smaller number. Published comparisons of waist sites in the same people have found differences of up to 6.9 cm in men and 10.1 cm between the extreme sites in women, so using the wrong one is not a rounding error. Take it at the end of a normal breath out, with the tape snug and horizontal and not compressing the skin.

Hip circumference (cm)Used only by the women’s Navy equation, where it is added to the waist. The published protocol for the hip in the original report is the circumference taken just below the gluteal fold. The field is still required when the men’s equations are selected, because the page computes both equation sets and shows you the one you did not pick; it has no effect on the men’s figure. Hip measurements are the least reproducible of the three girths, both because the landmark is harder to find and because clothing interferes more.

Biceps skinfold (mm)Skinfold method only. Vertical fold on the front of the upper arm, over the belly of the biceps, midway between shoulder and elbow, arm hanging relaxed. All four skinfolds need a calibrated caliper that applies a known pressure and, realistically, somebody else to take them — the subscapular and suprailiac sites cannot be reached properly on yourself and a fold pinched by hand is not the same quantity. Take each site two or three times and use the median rather than the first reading. Published training studies put the between-observer error on a single site at one to two millimetres even among trained measurers, which propagates into the estimate.

Triceps skinfold (mm)Skinfold method only. Vertical fold on the back of the upper arm, midway between the tip of the shoulder blade’s acromion and the point of the elbow, arm hanging relaxed. This is the most reproducible of the four sites and the one most often used alone, though a single site is a considerably worse predictor than the sum of four.

Subscapular skinfold (mm)Skinfold method only. Fold just below the inferior angle of the shoulder blade, taken at about 45 degrees to the horizontal, running down and out along the natural cleavage of the skin. Needs a second person. Of the four sites this one carries the most weight in distinguishing central from peripheral fat, which is part of why the four-site sum predicts better than the arm sites alone.

Suprailiac skinfold (mm)Skinfold method only. Fold just above the iliac crest, in the midaxillary line, running slightly downward and forward along the natural fold line. Protocols for this site differ more between textbooks than any of the other three, and the fold is both larger and more compressible here, so it is the single largest contributor to disagreement between two measurers. If one site has to be repeated, repeat this one.

20.7%Example

Navy method, men’s equations, 35 years, 175 cm, 75 kg, neck 38 cm, waist 90 cm, hip 100 cm, skinfolds 8 / 12 / 14 / 16 mm

Two regressions on body density, one conversion from density to a percentage, and the published scatter around each

Navy, men (cm): D = 1.0324 − 0.19077·log₁₀(waist − neck) + 0.15456·log₁₀(height) · Navy, women (cm): D = 1.29579 − 0.35004·log₁₀(waist + hip − neck) + 0.22100·log₁₀(height) · Skinfold: D = c − m·log₁₀(biceps + triceps + subscapular + suprailiac) · Siri: %fat = 495/D − 450 · Brozek: %fat = 457/D − 414.2 · 95% interval = %fat ± 1.96 × SEE, with SEE 3.52 points for men and 3.72 for women

D: whole-body density in g/cc. Every figure on this page is a prediction of density first and a percentage second, which is why the choice of density-to-percentage conversion matters and is shown
waist: a different site in each Navy equation. The men’s was fitted on the circumference at the umbilicus (Abdomen II in the source); the women’s on the circumference at minimal abdominal width, roughly midway between the lower sternum and the umbilicus (Abdomen I). Published comparisons of waist sites in the same people differ by up to 6.9 cm in men and 10.1 cm in women between the extremes
neck, hip: neck just below the larynx with the tape sloping slightly downward at the front; hip just below the gluteal fold. Both as defined in the original reports. The neck is subtracted, so an error in it moves the answer opposite to the same error in the waist
c, m: the Durnin and Womersley constants, by sex and age band. Men: 1.1620/0.0630 (17–19), 1.1631/0.0632 (20–29), 1.1422/0.0544 (30–39), 1.1620/0.0700 (40–49), 1.1715/0.0779 (50 and over). Women: 1.1549/0.0678, 1.1599/0.0717, 1.1423/0.0632, 1.1333/0.0612, 1.1339/0.0645 for the same bands. The men’s 30–39 pair is out of line with its neighbours, which is why the step at 40 is the largest of the three
495 / D − 450: Siri’s 1961 two-compartment conversion, equivalently 100 × (4.95/D − 4.50). Used by both source papers, so it is the conversion the published standard errors belong to. It assumes fixed densities for fat and for everything else, in everybody
SEE: the standard error of estimate from each derivation sample: 0.00791 g/cc or 3.52 percentage points for the men’s Navy equation (n = 602, R = 0.90) and 0.00796 g/cc or 3.72 points for the women’s (n = 214, R = 0.85). Cross-validation on independent samples gave 2.70 points for men and 4.04 and 4.36 for women. The interval printed above uses the derivation figures
what the interval is NOT: it is the scatter of the regression about underwater weighing in the 1980s derivation samples. It does not include measurement error on your own tape or caliper, and it does not include the bias these equations show against DXA, which independent studies have put at anything from 2.3 to 6.0 percentage points and in opposite directions for women. The true uncertainty is wider than the interval

Worked example

Navy method, men's equations, 35 years, 175 cm, 75 kg, neck 38 cm, waist 90 cm, hip 100 cm, skinfolds 8 / 12 / 14 / 16 mm
The girth term first, because it is the whole equation. Waist minus neck = 90 − 38 = 52 cm, and log₁₀(52) = 1.71600. log₁₀(175) = 2.24304. So D = 1.0324 − 0.19077 × 1.71600 + 0.15456 × 2.24304 = 1.0324 − 0.32735 + 0.34667 = 1.051722 g/cc, and Siri gives 495 / 1.051722 − 450 = 20.66%, printed as 20.7.
What that figure is worth. The men's equation was fitted on 602 naval personnel with a standard error of estimate of 3.52 percentage points, so the 95% interval is 20.7 ± 6.9, which is 13.8% to 27.6%. On a 75 kg body that span is 10.3 kg of fat. Every one of those values is consistent with these three tape measurements. A page that prints 20.7 and stops has told the reader something true and left out the only thing that makes it interpretable.
The same body on the skinfold method. The four sites total 8 + 12 + 14 + 16 = 50 mm, log₁₀(50) = 1.69897. At 35 the men's constants are 1.1422 and 0.0544, so D = 1.1422 − 0.0544 × 1.69897 = 1.049776 g/cc and Siri gives 21.53%. The two methods are 0.87 points apart here, which is close agreement — and close agreement between two regressions fitted against the same reference is not evidence that either is right.
The birthday step, which is the skinfold method's own artefact. With the same 50 mm total, the 40 to 49 constants are 1.1620 and 0.0700, giving D = 1.043072 and 24.56%. So this reader's skinfold estimate rises by 3.03 percentage points on their fortieth birthday, with nothing measured changing. Across the bands from 29 to 50 the same 50 mm gives 18.9%, 21.5%, 24.6% and 26.4% — a 7.5-point range from age alone. The Navy equations ignore age entirely, which is a different error rather than a smaller one.
Which Navy equation, which is the question nobody asks. The 1984 reports predict density in centimetres, which is what the headline used. The reformulated version in the US Department of Defense instruction predicts the percentage directly from inches: 86.010 × log₁₀(52/2.54) − 70.041 × log₁₀(175/2.54) + 36.76 = 20.78%. Here the two forms agree to 0.12 points. Swept across a realistic range of heights, necks and waists they diverge by up to 2.0 points for men and 3.7 for women, and almost no site states which it implemented.
And the conversion from density to percentage. Brozek's equation on the same density of 1.051722 gives 457 / 1.051722 − 414.2 = 20.33%, 0.33 points below Siri. The two cross over at a density of about 1.063 g/cc, so the sign of the difference changes depending on how lean the body is: at a density of 1.02 Siri reads 35.3% and Brozek 33.8%, and at 1.08 Siri reads 8.3% and Brozek 8.9%.
What the external validations do to all of this. Against DXA in 609 US Marines the circumference method underestimated men's body fat by 2.6 points, which would put this reader at 23.3%; in 1,407 military trainees it underestimated by 6.0 points, which would put them at 26.7%. The two studies disagree. The derivation interval of 13.8% to 27.6% does not include that disagreement, because it is the scatter of a regression about its own reference, so the honest reading of 20.7% is a figure somewhere in the low-to-middle twenties with the direction of the remaining bias unknown.
What this page does not print. No category, no range, no target. A body fat percentage with a "healthy range" beside it is the most common form of this page on the internet, and the ranges in circulation are either from organisations whose licences do not permit commercial reproduction or from sources whose own numbers could not be verified for this build. A reference distribution does exist in the public domain — the NCHS report on DXA body composition in the US population — and is cited below rather than reproduced, because its tables could not be read to the accuracy this project requires. Given a method whose error is several percentage points wide, a band boundary drawn across it would be a line through noise.

The two methods, their derivation samples, and the error each one published

The error column is the reason this page exists in the shape it does. A standard error of estimate of 3.52 percentage points means a 95% interval roughly 14 points wide, which on a 75 kg body is about ten kilograms of fat. For the women’s Navy equation the cross-validation errors are larger still, at 4.04 and 4.36 points, and the derivation sample was a third the size of the men’s. The last row is an admission: the per-age-band standard errors from the 1974 skinfold paper could not be obtained for this build, so the page prints an interval for the Navy method and not for the skinfold method, rather than printing an interval whose source it cannot name.
Method and form	Derivation sample	Reference method	Multiple R	Standard error of estimate	Cross-validation
Navy circumference, men, density form in cm	602 male US Navy personnel, 18–56	underwater weighing, Siri conversion	0.90	0.00791 g/cc = 3.52 points of fat	100 men: r 0.90, SEE 2.70 points
Navy circumference, women, density form in cm	214 female US Navy personnel, 18–44 (mean 26.5)	underwater weighing, Siri conversion	0.85	0.00796 g/cc = 3.72 points of fat	80 USN women: r 0.87, SEE 4.04. 66 Canadian Forces women: r 0.80, SEE 4.36
Navy circumference, reformulated direct-percentage form in inches	594 men; 202 women (the reformulation report’s own counts)	as above	0.903 men, 0.856 women	3.52 points men, 3.61 points women	stated to give the same answer as the density form to the nearest whole percent
Durnin and Womersley, four skinfolds	209 men and 272 women, 16–72, Scotland	underwater weighing, Siri conversion	not reported per band in a form this page could verify	the primary per-band standard errors could not be obtained; see the licensing and provenance note below	none in the original paper

What the same four skinfolds give at different ages, because the published constants band age

Read down either of the percentage columns. The skinfolds are identical in every row; only the birthday differs. For men the same 50 mm total gives 18.9% at 29 and 26.4% at 55, a range of 7.5 percentage points, with the single largest jump — 3.03 points — falling on the fortieth birthday, because the 30 to 39 constants are out of line with their neighbours. For women the same 60 mm gives 29.5% at 29 and 35.7% at 55. There is a real effect underneath: at a given subcutaneous fat thickness, older adults do carry proportionally more fat. The published representation of it is a staircase and the thing it represents is not, so a reader within a year or two of a band edge should read both rows.
Age band	Men’s constants (c, m)	Men at 50 mm total	Women’s constants (c, m)	Women at 60 mm total
17–19	1.1620, 0.0630	19.2%	1.1549, 0.0678	28.6%
20–29	1.1631, 0.0632	18.9%	1.1599, 0.0717	29.5%
30–39	1.1422, 0.0544	21.5%	1.1423, 0.0632	30.6%
40–49	1.1620, 0.0700	24.6%	1.1333, 0.0612	33.2%
50 and over	1.1715, 0.0779	26.4%	1.1339, 0.0645	35.7%

What independent comparisons against DXA have found, and where they disagree

Two studies, both large, both against DXA, and they do not agree. They differ on the magnitude of the bias by a factor of more than two, and in women they differ on its direction. That is the state of the evidence and this page prints it rather than picking the flattering one. Two further points follow. The first is that both samples are military, young and fit, so neither is a general population — which is the same objection that applies to the equations themselves. The second is the more useful: if a method cannot reliably detect a four-point change in body fat over eight weeks of hard training, it cannot reliably detect whatever change a reader is hoping to see between one month and the next. For tracking change over time, the sum of the raw skinfolds in millimetres, or the waist in centimetres, is a better quantity than any percentage derived from them, because it drops the regression and its error.
Study	Sample	Bias in men	Bias in women	What else it found
US Marine Corps body composition survey (2022)	609 Marines: 430 men, 179 women, 18–57	circumference UNDERestimated DXA by 2.6 ± 3.7 points (age 30 or under) and 2.5 ± 3.7 (over 30)	circumference OVERestimated DXA by 2.3 ± 4.3 points (30 or under) and 1.3 ± 4.8 (over 30)	lean people were read high and people with more fat were read low. The authors concluded the method is serviceable for classification but not suitable where quantitative body composition is needed
Eight weeks of military training (2023)	1,407 trainees: 926 men, 481 women	circumference underestimated DXA by 6.0 ± 3.5 points	circumference underestimated DXA by 6.0 ± 4.4 points	it failed to detect change: DXA measured a 4.0 ± 2.4 point fall in the women’s body fat where circumference measured 0.0 ± 3.3. Only 56% of the women were correctly identified as having changed, against 83% of the men

Where a reference distribution for body fat percentage can be found, and why none is reproduced here

Printing nothing is a worse page than printing a well-sourced distribution and a better page than printing a borrowed one. Two considerations make the absence easier to defend here than it would be elsewhere. The first is arithmetic: with a method whose 95% interval is fourteen percentage points wide, a boundary drawn anywhere across it is a line through noise, and a reader straddling it would be told two different things by two equally valid readings of their own tape. The second is the rule this whole group of pages is built on. A body fat percentage with a target range beside it is the exact shape of output that someone anxious about their body will read as an instruction, and it is the shape with the least evidence behind it of anything on this page.
Source	What it contains	Licence position	Used here?
NCHS, Vital and Health Statistics Series 11 No. 250 (2010)	DXA body composition for the US population aged 8 and over, NHANES 1999–2004, 22,010 individuals, with percentile tables by sex, age and ethnicity	US Government work, explicitly public domain: “All material appearing in this report is in the public domain and may be reproduced or copied without permission”	cited, not reproduced. The report’s own tables could not be read to the accuracy this project requires for a published figure, and two attempts at extracting them disagreed materially
Kelly, Wilson and Heymsfield, PLoS ONE 2009;4(9):e7038	DXA body composition reference values from the same NHANES data, with percentile tables by sex and ethnicity in supplementary tables	CC BY, which permits commercial reuse with attribution	cited, not reproduced. The percentile values are in supplementary files that could not be retrieved for this build
Fitness and exercise-science organisations’ body fat classification tables	the familiar grids that sort a percentage into named tiers	proprietary; these tables are the organisations’ own published recommendations and are not licensed for reproduction on an advertising-supported site	not used, and not paraphrased either, which would be the same thing wearing a disguise
The body fat criterion used in BMI accuracy studies (over 25% in men, over 35% in women)	a pair of threshold figures	the provenance of that pair is a WHO figure, and WHO publications are CC BY-NC-SA 3.0 IGO; the NonCommercial term applies to this site	not adopted as a threshold on this page. The studies that used it are cited for their own results

What a regression on girths or skinfolds can and cannot tell you about the inside of a body, and why the error bar is the headline

Nothing on this page measures body fat. Both methods here were built the same way: somebody measured a few hundred people twice — once by submerging them in water to get their whole-body density, and once with a tape measure or a caliper — and then fitted a line from the cheap measurement to the expensive one. What a calculator returns is a point on that line. It is the average body fat of people whose tape measurements look like yours, in the sample the line was fitted on, and the scatter of real people around that line is published and large. For the US Navy men’s equation the standard error of estimate is 3.52 percentage points, which puts a 95% interval roughly fourteen points wide around any figure it produces. On a 75 kg body that is ten kilograms of fat. Every value in that interval is consistent with the three numbers you typed.

That interval is the reason this page shows two methods rather than one. A reader given a single figure of 18.4% has no way to know it came out of a fourteen-point window, and no way to know that a different published equation, applied to the same measurements, would have said 21.9%. Showing both, with the gap between them, converts an invisible uncertainty into a visible one. It is not a perfect proxy — two regressions fitted against the same reference can agree closely and both be wrong, which is what the DXA comparisons suggest actually happens — but a reader who can see that two published methods disagree by three points will not read the first decimal place of either as meaningful, and that is most of the battle.

Why the US Navy equations, and what their origin means. They are the most widely implemented circumference method in the world, they need nothing but a tape measure, and their primary sources are two 1984 US Government technical reports that can be read in full, which means the coefficients on this page were taken from the derivation rather than from somebody’s copy of somebody’s copy. That matters more than it sounds: the detail that the men’s equation uses the waist at the umbilicus and the women’s uses the waist at its narrowest point is in those reports and is dropped by almost every online implementation, which simply asks for “waist”. Since published comparisons of waist sites in the same people differ by up to 6.9 cm in men and 10.1 cm in women, and one centimetre of girth is worth about 0.7 percentage points at an average male build, measuring at the wrong site is a larger error than the regression’s own.

The generalisation problem is real and runs in a specific direction. The men’s equation was fitted on 602 serving US Navy personnel aged 18 to 56 and the women’s on 214 aged 18 to 44, and those are not samples of the general population: they were selected for fitness, they were young, they were overwhelmingly American, and nobody in them was outside the range of body composition that military service permits. Two consequences follow for a reader who is not a young American sailor. The equations have no age term at all, so they implicitly assume the relationship between girths and fat that held in a 1980s naval cohort, and that relationship does shift with age as lean mass is lost. And the error is not uniform across body sizes: the Marine Corps study found lean people read high and people carrying more fat read low, which means the method compresses the range — it will tend to tell everybody they are nearer the middle than they are.

Why Durnin and Womersley rather than Jackson and Pollock. Both are legitimate and both are in wide use, and the choice is a trade-off that should be stated rather than hidden. Durnin and Womersley’s 1974 study measured 209 men and 272 women aged 16 to 72 in Scotland, and their four-site equation uses the same four skinfolds — biceps, triceps, subscapular, suprailiac — for both sexes, with a different pair of constants for each sex and age band. Jackson and Pollock’s generalised equations, published for men in 1978 and for women in 1980, use a quadratic in the skinfold sum plus age as a continuous term, which avoids the banding artefact entirely, but they use different sites for men and women and their samples were younger and more athletic. Three things decided it here. The same four sites for both sexes means one protocol to describe and one fewer way for a reader to use the wrong landmark. The age range of 16 to 72 is far closer to a general adult population than Jackson and Pollock’s samples. And the published form is two constants per cell, which is a far smaller surface for a transcription error than four coefficients per sex, in a project where the only acceptable source for a coefficient is one that can be checked. The cost of that choice is the staircase: the estimate steps by two to three percentage points at each band edge, and the page prints the size of the step for the figures you entered rather than pretending it is not there.

The conversion from density to percentage is a second set of assumptions, stacked on the first. Both methods predict whole-body density, and turning a density into a proportion of fat requires assuming a density for fat tissue and a density for everything else, in everybody. Siri’s 1961 equation and Brozek’s 1963 equation make slightly different assumptions and cross over at a density of about 1.063 g/cc, so the sign of the difference between them depends on how lean the body is. This page uses Siri because that is what both source papers used, which means the published standard errors belong to it. But the two-compartment assumption is known to be wrong in a predictable direction for anybody whose bone mineral density is unusually high or low, which includes athletes, older adults and several clinical groups; a density-based estimate reads fat too high when the fat-free body is less dense than assumed. These are not refinements on top of a measurement. They are assumptions underneath an estimate.

What this page deliberately does not have is a category, a range or a target. That is the most consequential editorial decision on it, so here is the reasoning in full. A reference distribution for body fat percentage does exist in forms that could lawfully be reproduced on this site: the NCHS report on DXA body composition in the US population is a public-domain US Government work, and the PLoS ONE reference-values paper built on the same survey is CC BY. Neither could be read to the standard this project requires for a published number during this build — the NCHS tables resisted two extraction attempts that disagreed with each other materially, and the PLoS supplementary tables could not be retrieved. Both are cited below so a reader can go and look. The familiar classification grids from fitness organisations are a different matter: they are those organisations’ own recommendations, they are not licensed for reproduction on an advertising-supported site, and paraphrasing one to get around that would be the same breach in a different typeface. And even with a clean distribution in hand there is an arithmetic objection: with a 95% interval fourteen percentage points wide, a boundary drawn across it is a line through noise, and a reader sitting near one would be sorted into either of two categories by the ordinary variation in where a tape goes.

If what you want is to follow a change over time, use a different quantity. The most striking result in the validation literature is not the bias, it is the failure to detect change: over eight weeks of military training, DXA measured a four-point fall in body fat in 481 women, and the circumference method measured zero. The regression is the problem. Every step from tape to percentage adds error, and tracking a derived percentage month to month means tracking the noise of that derivation as well as the signal. The sum of the four skinfolds in millimetres, or the waist in centimetres measured the same way by the same person at the same time of day, is a better instrument for detecting change, precisely because it is not an estimate of anything. If you want a figure for the inside of your body with an error small enough to follow, the answer is DXA, and it is a scan rather than a sum.

Where the rest of this set fits. BMI needs two measurements and is the only body-size measure with decades of outcome evidence indexed on it, at the cost of not being able to tell muscle from fat at all. Waist-to-height ratio needs two measurements, has better cardiometabolic outcome evidence than BMI, and has the one cleanly licensed boundary in this set. Waist-to-hip ratio was the measure that beat BMI in the largest heart-attack case-control study ever run, and this site presents it without a threshold table for reasons that page sets out. The four pages answer the same underlying question four ways and they will not agree with each other; that disagreement is a more honest picture of what is actually known about a body from a tape measure than any single figure. For the related weights used in drug dosing rather than in self-assessment, lean body weight by the Janmahasatian equation is the counterpart in the medical set and is a size descriptor for sizing drug doses, not an estimate of anybody’s fat.

Frequently asked questions

Which of the two methods is more accurate?

There is no answer that holds for everybody, which is why the page computes both. What can be said is narrower. The Navy circumference method needs only a tape and can be done alone, and its error against underwater weighing in its own derivation sample was 3.52 percentage points for men and 3.72 for women; against DXA in independent military samples it has been found biased by 2.3 to 6.0 points, with the direction disagreeing between studies in women. The skinfold method is measuring something closer to the quantity of interest — the thickness of fat under the skin — but it needs a calibrated caliper and a second person, two of its four sites cannot be reached on yourself, and its published constants band age so the estimate steps by two to three points on three birthdays. If you must pick one, pick the one you can do consistently, because for anything you actually want to know the repeatability matters more than the calibration.

Why does my answer differ from another website’s, using the same measurements?

Usually one of four things, and this page prints all four so you can work out which. Which form of the Navy equation: the 1984 density equations in centimetres and the reformulated direct-percentage equations in inches are meant to be equivalent and diverge by up to 2.0 percentage points in men and 3.7 in women across a realistic range. Which conversion from density to percentage: Siri and Brozek differ by up to about one and a half points and the sign of the difference depends on how lean the body is. Which skinfold equation: Durnin and Womersley and Jackson and Pollock are different equations with different site counts and routinely differ by several points. And which waist site: the men’s and women’s Navy equations use different ones, and most sites just say “waist”. Almost no calculator states any of these, which is the real reason the internet cannot agree about anybody’s body fat.

Why is there no healthy range or category?

Three reasons, in order of weight. First, with a 95% interval fourteen percentage points wide, a category boundary drawn across it is a line through noise — a reader near the boundary would be sorted into either of two categories by the ordinary variation in where a tape goes, which makes the category an artefact of the measurement rather than a fact about them. Second, the classification grids in common circulation are the published recommendations of fitness and exercise-science organisations and are not licensed for reproduction on a site that carries advertising; paraphrasing one to get around that is the same breach in different words. Third, two reference distributions that could lawfully be used do exist — a public-domain NCHS report and a CC BY paper, both cited below — and neither could be read accurately enough for this build, so they are pointed at rather than quoted. A number with no range beside it is a less satisfying page and a more honest one.

Can I use this to track whether my body fat is falling?

Not reliably, and the evidence on that point is unusually clear. In 1,407 military trainees over eight weeks of training, DXA measured a 4.0 percentage point fall in body fat in the women; the circumference method measured 0.0. Only 56% of the women were correctly identified as having changed at all. Every step from the tape to the percentage adds error, so tracking a derived percentage over time means tracking the noise of the derivation as well as the signal. If you want to follow a change, follow a measurement rather than an estimate: the sum of your four skinfolds in millimetres, or your waist in centimetres, taken the same way by the same person at the same time of day. Those are smaller numbers with smaller errors and they move when something moves.

How much does getting the measurement wrong cost me?

More than the arithmetic does, which is the uncomfortable part. The rows above print it for your own figures: the effect of one centimetre on the Navy girth term, and of four millimetres on the four-site skinfold total. For a man at a waist of 90 cm and a neck of 38 cm, one centimetre on the waist is worth about 0.7 percentage points — so the up-to-6.9 cm spread between waist measurement sites found in published comparisons is worth nearly five points on its own, before any regression error enters. For skinfolds, published training studies put the between-observer error on a single site at one to two millimetres even among trained measurers, and the suprailiac site is the worst of the four. Take each site two or three times and use the median; take the waist at the end of a normal breath out with the tape horizontal and snug but not compressing.

Does the equation set control mean the equations are about my sex or my gender?

Neither, strictly. It selects which of two published regressions is applied, and those regressions were fitted on samples that the original studies recorded as male and female. The equations differ because the samples differed in average body composition and in where fat was distributed, not because of anything about identity, and for the Navy method the two equations additionally use different measurement sites — so choosing the wrong set means measuring the wrong place as well as using the wrong constants. If neither set obviously fits, computing both and reading the span between them is a more honest answer than a confident figure from either, and the page makes that easy because it always computes both methods and shows the one you did not select.

Is a lower body fat percentage better?

No, and this page will not imply it. Fat is structural and metabolically necessary: it insulates, it cushions organs, it stores energy, and it is involved in hormone production, which is why extremely low body fat is associated with periods stopping, bone loss and impaired immune function rather than with health. There is no direction of travel on this page and nothing on it improves by making the number smaller. The figure above is an estimate, with an error of several percentage points, of one property of a body, produced by a regression fitted on a few hundred sailors in the 1980s. If the number matters to you more than the error bar around it does, that is worth mentioning to a doctor.

What about bioimpedance scales, the ones that send a current through your feet?

They are a third method with a fourth set of problems and they are not on this page because the manufacturers do not publish their equations. A bioimpedance device measures the resistance of the body to a small current and then applies a proprietary regression to turn that into a percentage, so the coefficients cannot be checked, the derivation sample is unstated and the error is whatever the manufacturer says it is. They are also strongly affected by hydration, by how recently you ate or exercised, and by the time of day, which is why the same scale can read two points apart within a few hours. The methods here are no more accurate, but their equations, their samples and their published errors are all on this page and can be argued with.

Related calculators

References

Hodgdon JA, Beckett MB. Prediction of percent body fat for U.S. Navy men from body circumferences and height. Naval Health Research Center Report No. 84-11, San Diego, March 1984 (DTIC AD-A143 890). A US Government technical report; the primary source for the men’s equation implemented here. Body density = 1.0324 − 0.19077 × log₁₀(Abdomen II circumference − neck circumference) + 0.15456 × log₁₀(height), with all measurements in CENTIMETRES; n = 602 male naval personnel aged 18–56; multiple R = 0.90; standard error of estimate 0.00791 g/cc, equivalent to 3.52 percentage points of fat; cross-validation on 100 further men gave r = 0.90 and a standard error of 2.70 points. Body composition was determined by underwater weighing and density converted with Siri’s equation. The site definitions reproduced on this page are the report’s: neck just inferior to the larynx with the tape sloping slightly downward to the front, Abdomen II at the level of the umbilicus. PROVENANCE NOTE: the sign on the height coefficient was checked explicitly, because the text extraction of this report renders it ambiguously; it is positive, which is required both by physical sense and by numerical agreement with the inch-coefficient form.
Hodgdon JA, Beckett MB. Prediction of percent body fat for U.S. Navy women from body circumferences and height. Naval Health Research Center Report No. 84-29, San Diego, June 1984 (DTIC AD-A146 456). The primary source for the women’s equation implemented here. Body density = 1.29579 − 0.35004 × log₁₀(Abdomen I + hip − neck) + 0.22100 × log₁₀(height), measurements in CENTIMETRES; n = 214 female naval personnel aged 18–44, mean 26.5; multiple R = 0.85; standard error of estimate 0.00796 g/cc, equivalent to 3.72 percentage points; cross-validated on 80 US Navy women (r = 0.87, SEE 4.04 points) and 66 Canadian Forces women (r = 0.80, SEE 4.36 points). Site definitions as published: neck as above; Abdomen I at the level of minimal abdominal width, approximately midway between the xiphoid and the umbilicus; hip just inferior to the gluteal fold. The difference between the men’s abdominal site and the women’s is stated prominently on this page because it is routinely dropped elsewhere.
Hodgdon JA. Body composition in the military services: standards and methods. Naval Health Research Center (DTIC AD-A370 158). Source of the reformulated direct-percentage equations in INCHES also computed on this page: men, %fat = 86.010 × log₁₀(abdomen II − neck) − 70.041 × log₁₀(height) + 36.76, n = 594, R = 0.903, SEE 3.52 points; women, %fat = 163.205 × log₁₀(abdomen I + hip − neck) − 97.684 × log₁₀(height) − 78.387, n = 202, R = 0.856, SEE 3.61 points. The report states the reformulated equations give virtually the same result as the originals when rounded to the nearest whole percent; swept over a realistic range of body dimensions for this page they diverge by up to 2.00 percentage points in men and 3.75 in women, which is stated rather than smoothed over. Note that the sample counts and the women’s standard error differ slightly from the 1984 derivation reports.
Siri WE. Body composition from fluid spaces and density: analysis of methods. In: Brozek J, Henschel A, eds. Techniques for Measuring Body Composition. Washington DC: National Academy of Sciences, 1961:223–44. The conversion from density to percentage used throughout this page: %fat = 100 × (4.95/D − 4.50), equivalently 495/D − 450. Used because it is the conversion both source papers used, so the published standard errors belong to it. A two-compartment model that assumes fixed densities for fat and fat-free tissue in everybody.
Brozek J, Grande F, Anderson JT, Keys A. Densitometric analysis of body composition: revision of some quantitative assumptions. Ann N Y Acad Sci 1963;110:113–40. Source of the alternative conversion printed in the rows above, %fat = 457/D − 414.2. Computed and shown because the choice between the two conversions moves the answer, by up to about 1.5 percentage points at the high-fat end, and because the two cross over at a density of approximately 1.063 g/cc so the sign of the difference is not constant. A conversion between two quantities is an arithmetic identity given its assumptions and is nobody’s property.
Durnin JVGA, Womersley J. Body fat assessed from total body density and its estimation from skinfold thickness: measurements on 481 men and women aged from 16 to 72 years. Br J Nutr 1974;32(1):77–97. doi:10.1079/BJN19740060. The source of the skinfold equations implemented here: body density = c − m × log₁₀(sum of biceps, triceps, subscapular and suprailiac skinfolds in mm), with c and m by sex and age band. 209 men and 272 women aged 16 to 72, body density by underwater weighing, body fat ranging from 5–50% in the men and 10–61% in the women; the authors established that the logarithmic transformation was required to linearise the relationship, and fitted separate equations by sex and age group. PROVENANCE NOTE, stated because it is part of what this page had to establish: the primary table could not be obtained in full for this build, so the twenty constants used here were taken from two independent compilations that agree with each other exactly on every cell, and the per-band standard errors could not be verified at all — which is why this page prints a published interval for the Navy method and does not print one for the skinfold method. The constants used are given in full in the formula block above so that anyone with the paper can check them against it.
Harty PS, Friedl KE, Nindl BC and colleagues. Circumference-based predictions of body fat revisited: preliminary results from a US Marine Corps body composition survey. Front Physiol 2022;13:868627. 609 US Marines (430 men, 179 women) aged 18–57, circumference-estimated body fat against DXA. In men the circumference method underestimated DXA by 2.6 ± 3.7 percentage points at age 30 or under and 2.5 ± 3.7 above 30; in women it overestimated by 2.3 ± 4.3 and 1.3 ± 4.8. Lean individuals were overestimated and higher-fat individuals underestimated. The authors’ conclusion, quoted on this page in substance, is that the method is a field-expedient classification tool and is not suitable where quantitative body composition data is required.
Foulis SA, Friedl KE, Spiering BA and colleagues. Body composition changes during 8 weeks of military training are not accurately captured by circumference-based assessments. Front Physiol 2023;14:1183836. 1,407 trainees (926 men, 481 women) against DXA. Circumferences underestimated DXA body fat by 6.0 ± 4.4 percentage points in women and 6.0 ± 3.5 in men at baseline, and failed to detect change: DXA measured a 4.0 ± 2.4 point fall in the women where circumferences measured 0.0 ± 3.3 (p = 0.86), with 56% of women correctly identified as having changed against 83% of men. Post-training agreement was r² = 0.36, SEE 3.56% in women and r² = 0.54, SEE 3.28% in men. Cited on this page for the bias figures and, more importantly, for the failure to track change, which is the result that should govern how anybody uses a figure like the one above over time.
Borrud LG, Flegal KM, Looker AC, Everhart JE, Harris TB, Shepherd JA. Body composition data for individuals 8 years of age and older: U.S. population, 1999–2004. National Center for Health Statistics, Vital and Health Statistics Series 11, No. 250, February 2010. DXA body composition on 22,010 individuals from NHANES, with percentile tables for percentage body fat by sex, age and ethnicity. A US Government work, explicitly in the public domain: “All material appearing in this report is in the public domain and may be reproduced or copied without permission.” CITED AND NOT REPRODUCED. This is where a cleanly licensed reference distribution for body fat percentage exists; its tables could not be read to the accuracy this project requires during this build, and two extraction attempts disagreed materially, so the page points at the source rather than printing figures it cannot stand behind.
Kelly TL, Wilson KE, Heymsfield SB. Dual energy X-ray absorptiometry body composition reference values from NHANES. PLoS ONE 2009;4(9):e7038. doi:10.1371/journal.pone.0007038. Licensed CC BY, which permits commercial reuse with attribution. A second cleanly licensed route to a reference distribution for percentage body fat from the same survey. Cited and not reproduced: the percentile values are held in supplementary tables that could not be retrieved for this build.
LICENSING POSITION TAKEN FOR THIS PAGE, recorded deliberately. The two Hodgdon and Beckett reports and the US Department of Defense reformulation are US Government works. The Durnin and Womersley, Siri, Brozek, Marine Corps and military-training papers are cited with attribution for their equations and their results, which is ordinary scholarly use of methods and findings. No body fat classification table from any fitness or exercise-science organisation is reproduced or paraphrased anywhere on this page, because those tables are those organisations’ own published recommendations and this site carries advertising. No WHO material is used: in particular the body fat criterion of over 25% in men and over 35% in women, which appears in several of the validation studies cited here, traces to a WHO figure, and this page cites those studies’ results without adopting that criterion as a threshold. No NICE material is used. The single alternative to printing a borrowed range — printing no range — is what this page does, and the table above sets out where a reader can find a properly licensed distribution for themselves.

CalcEngines health calculators are for education and for checking arithmetic that has already been decided elsewhere. They are not medical advice, they do not decide what to give, and they do not replace the judgement of a doctor, nurse, midwife or dietitian who knows the person in front of them. Every figure depends on the values you enter and on the assumptions stated on the page — check it against the prescription, the product label and your local policy before acting on it.