Best Way to Measure Body Fat: Pick the Method You Can Repeat
Every home method turns a handful of measurements into one percentage, and every one of them is limited by the same thing: how closely you can repeat the measurements you take. This page turns each method into a single number — its repeat spread, in body fat points — and then asks the only question that decides anything: how small a change can it see? At the defaults below, a tape read to half a centimetre carries 0.57 points, a caliper read to one millimetre carries 0.52, and a smart scale with 1.5 points of daily noise carries 4.16. A caliper only beats a tape if you can repeat a pinch to within 1.09 mm. Everything here is arithmetic this page performs on published equations. Nothing on it ranks a product, and nothing on it compares any of these methods against a laboratory reference.
Caliper wins, at a repeat spread of 0.52 points versus 0.57 for tape — a margin of 9.4%. That is close enough that one extra reading per site flips it: the loser needs only 2 readings to tie. For the tape to match that you would need every circumference repeatable to 0.46 cm; for the caliper to lose you would have to pinch worse than 1.09 mm per site.
To resolve a 1.00 point change you are allowed 0.89 cm of slop on every circumference and 1.92 mm on every pinch — and height may be off by 5.8 cm before it matters. At 1 reading per site you can see 1.44 points today; reaching 1.00 takes 3 readings per site with the caliper, 3 with the tape, and 18 with your device.
At 0.50 points a month, a real change takes 2.9 months before the caliper can call it — and 8.3 months before your device can. Your tape estimate reads 17.0%, the three-site caliper reads 13.2%, and the BMI equation predicts 22.2% for your weight, age and sex.
Where your tape error comes from: waist 48.8%, neck 48.8%, height 2.3%. Halving the biggest one buys 20.4% of your spread; halving the smallest buys 0.9%.
The short answer
There is no best method in the abstract. There is a best method for you, and it is decided by one number you can measure in ten minutes: how closely you repeat yourself. Everything else — the reputation of the method, the number of decimal places, the price of the instrument — is second order against that.
- Repeatability is the whole game for tracking. A method with a fixed offset still tracks you perfectly; a method that jumps around by two points hides a two-point change. The tables below measure exactly that, and nothing else.
- A caliper beats a tape only if you can pinch to about one millimetre. At half a centimetre of tape slop the break-even is 1.09 mm per site for a man and 0.80 mm for a woman. Most people who buy a caliper never get there, and would have been better off with the tape.
- The neck is half of a man's tape error. Waist and neck carry 48.8 percent each of the variance at typical slop, and almost nobody practises the neck. Halving your neck slop buys 20.4 percent of your total spread, which is more than any equipment purchase would.
- The answer flips with the body, not with the method. The tape gets better as you get larger, because the Navy equation divides by your waist-minus-neck. On the eight computed bodies below the caliper wins on lean and average men, and the tape wins on every woman and on the largest man.
If you want one instruction: take the method you already own, take two readings at every site every time, and write down the number. Two readings per site buys more than switching methods, at every slop level in the tables below.
Computed: what “repeatable” means in centimetres
This is the inversion most people never do. Instead of asking how much error a given slop produces, ask how much slop you are allowed if you want the answer to hold still to within a chosen number of points. Differentiating the Navy equation gives 37.35 ÷ (waist − neck) points per centimetre for men and 70.88 ÷ (waist + hip − neck) for women, so the allowance is simply the target divided by that.
Two things fall out. The first is that the allowance is brutally small on a lean body: a man whose waist exceeds his neck by 20 cm has to place every circumference to within two millimetres to hold his estimate to half a point. The same tape work that is fine at 60 cm is hopeless at 20, which is why lean people conclude that these equations “do not work on them” — the equation is fine, the allowance has simply shrunk below what a hand can do.
The second is that height barely matters here. The columns above already charge you for every circumference wobbling at once, which is why they are tighter than the single-input number. Height alone enters at 30.42 ÷ height for men and 42.42 ÷ height for women, so a one point allowance is 5.8 cm at 175 cm and 3.8 cm at 162 cm. You do not need a stadiometer; you need a tape you place the same way twice.
The same inversion for the caliper. Jackson-Pollock three-site sums three folds, so three independent wobbles combine as √3, and the allowance per site is the target divided by √3 times the per-millimetre sensitivity:
Note that the caliper allowance moves in the opposite direction from the tape: it gets looser as the folds get thicker, because the density curve flattens out. A tape punishes lean bodies, a caliper punishes lean pinches. That single difference is what makes the comparison below interesting instead of obvious.
Computed: where your error actually comes from
Shares are computed as each input's squared contribution over the total variance, at half a centimetre of circumference slop, half a centimetre of height error and one millimetre per pinch. The right-hand column is what you gain by halving that one input's slop and changing nothing else.
The practical reading is uncomfortable. For a man, the neck — the measurement nobody rehearses — carries exactly as much of the error as the waist, and halving either one buys a fifth of the total spread. For a woman the tape error is spread evenly over three circumferences, so there is no single place to improve: each halving buys only 12.3 percent, and the honest fix is to take more readings rather than to try harder on one site.
The BMI equation is the odd one out. Its two inputs are the easiest things you own to repeat, so its repeat spread is the smallest number on this page — 0.211 points for the reference man. That is not a recommendation. That equation never looks at your body at all; it is arithmetic on your weight, height, age and sex, and its repeatability is small precisely because it is measuring almost nothing. A number that is easy to repeat and impossible to be wrong about in the short term can still be wrong by a wide margin for you personally, and this page has no way to quantify that margin.
Computed: the break-even — when does a caliper beat a tape
Set the two spreads equal and solve for the pinch. Because averaging divides both sides by the same √n, the break-even does not depend on how many readings you take — that is the useful part. At half a centimetre of height error:
A millimetre is a small distance. It is roughly the thickness of a credit card, and it is the entire margin by which a caliper at typical pinch quality beats a tape at typical tape quality for a man. Once your tape work is good — 0.3 cm — the caliper has to be repeatable to two thirds of a millimetre just to draw. For women the bar is tighter still, because the Navy equation's denominator is the sum of three circumferences and therefore large: the tape wins at every realistic pinch quality until your tape work degrades past about a centimetre.
The same comparison run the other way gives the crossover in body size. Holding the slop at half a centimetre and one millimetre, and solving for the body where the two spreads are equal:
That is the structural reason the two methods keep switching places in the profiles below, and it is not something you can read off the instrument. A tape is a better instrument on a large body and a worse one on a lean body, because the number it computes is a ratio of two circumferences that grow together. A caliper is the reverse.
Computed: what each method can see, and what it costs to see more
Two sessions, 95 percent confidence, two-sided: the smallest change you may call real is 1.96·√2·σ, and averaging n readings at every site divides it by √n. At the reference man and woman, with half a centimetre of tape slop, one millimetre per pinch, 0.3 kg of weight spread, half a centimetre of height error and 1.5 points of device noise:
The first column is the one to read. A tape and a caliper sit within nine percent of each other on the reference man, and the fix for the loser is trivial: because both shrink as 1/√n, the loser needs only (0.569 ÷ 0.520)² = 1.2 — that is, two readings — to erase the gap entirely. Twenty seconds of extra work beats every equipment decision on this page.
The device column is the one people misread. A smart scale contributes its noise on top of your body's own, and unlike a tape it gives you no way to inspect where the number came from. To make it resolve a single body fat point you would need eighteen readings per comparison, which at daily weighing is over five weeks of data on each side of the comparison. That is not an argument against owning one; it is an argument against reading its daily number as if it were a measurement.
Computed: how long before a change is real
The same threshold read against time. If your body fat is genuinely moving at r points a month, you need threshold ÷ r months before a two-session comparison can see it. Reference man, at three rates and two readings-per-site settings:
Read the first row slowly. At a quarter of a point a month — a perfectly respectable rate — your tape cannot see anything you have done for more than half a year, and three readings per site cut that to three and a half months. Most people who conclude they have stalled have simply measured too soon, and no change of method would have helped them. The instrument was never the constraint; the calendar was.
The corollary is that the useful measurement interval is not the one you feel like taking. It is the threshold divided by your expected rate, rounded up, and it is worth writing on the calendar rather than deciding week to week.
Computed: eight bodies, the winner each time
Same slop assumptions throughout — half a centimetre per circumference, one millimetre per pinch, half a centimetre of height, 1.5 points of device noise, one reading per site. The winner is the smallest repeat spread.
The split is three to one by sex, and it is produced entirely by the two denominators: the Navy equation divides by a circumference difference or sum that is large on women, so women get more tape for the same slop, while the three-site equation is slightly more sensitive per millimetre on women. Notice also that the gap between the two methods is small everywhere — between 6 and 25 percent — which is the strongest argument on this page for spending your effort on repeats rather than on choosing.
The decision rule: which method for which job
Every row below is answered by the arithmetic above rather than by preference. Where the honest answer is “this page cannot tell you,” it says so.
Doing it: the steps that decide your number
The point of a protocol is not tidiness; every step below removes one of the σ terms in the equation at the top of this page. This is the short version — the longer at-home guide covers the same ground with more on each instrument.
- Mark the landmark once, with a washable pen, and measure to the mark every time. Moving the landmark by a centimetre is indistinguishable from your body changing, and it is the one error that never averages away because it is systematic within a session.
- Tape: waist at the level of the navel, neck just below the larynx, hip at the widest part of the seat — tape level all the way round, snug but not compressing. Take the reading at the same point in the breath every time; if you hold a breath in, the circumference you measure is not the one you measured last month.
- Caliper, men: a diagonal fold halfway between the nipple and the armpit crease; a vertical fold two centimetres beside the navel; a vertical fold halfway up the front of the thigh. Women: a vertical fold at the back of the upper arm, halfway between shoulder and elbow; a diagonal fold just above the hip bone along the natural crease; the same thigh fold.
- Lift skin and the fat under it, let the muscle drop away, and read at the same count every time. The fold compresses while you hold it, so a reading taken at one second and a reading taken at four are different measurements of the same body.
- Two readings at every site, every session, and rotate the sites so no site is measured twice in a row. Per the table above this is the cheapest improvement available and it beats every equipment change on the page.
- Fix the conditions: same time of day, same instrument, same person measuring, and not immediately after training, a sauna, a long flight or a salty meal. Each of those changes the thing you are measuring rather than the measurement.
- Measure on the schedule the arithmetic gives you, not when you feel like it. Threshold divided by your expected rate: three months at half a point a month, one and a half at one point a month.
Where this page's arithmetic stops being true
- Everything here is repeatability, not accuracy. The tables say how steadily a method reports its own number. They say nothing about how close that number is to your true body fat, and this page has no reference measurement to compare against, so it makes no claim of that kind anywhere.
- The published equations carry their own error, which is not in any of these numbers. The Navy, Jackson-Pollock and BMI equations were fitted on particular populations; how far they sit from you is a separate question this page cannot quantify. Every σ above is the part you add on top of that, not the total.
- Averaging shrinks random wobble and nothing else. A landmark you consistently place in the wrong spot, a tape that is stretched, or a caliper that is out of calibration produce the same wrong number every time, and the √n improvement does not touch them.
- Your slop figures are estimates until you measure them. The defaults are placeholders. To get the real ones: take three readings at each site in one session, twice in a week, and use the spread of those as σ. The calculator is only as good as the numbers you feed it.
- Independence is assumed.The quadrature sum treats each site's wobble as unrelated to the others. If one bad habit — pulling the tape too tight, say — affects every circumference in the same direction, the true spread is larger than what is shown here, not smaller.
- The BMI equation row is not a recommendation. It wins on repeatability because it contains almost no information about you. Small spread and small error are different things, and only the first is computed here.
- The device row is your figure, not a measured one. No device was tested for this page. The 1.5-point default is a placeholder for whatever your own scale does from day to day.
- None of this is medical advice. See the disclaimer.
Best way to measure body fat: questions
What is the best way to measure body fat at home?
The one you can repeat most closely, and for most people that is decided by their own hands rather than by the instrument. At half a centimetre of tape slop and one millimetre of pinch the two are within nine percent of each other on the reference man, and two readings per site erase the difference entirely. Pick one, mark your landmarks, take two readings at every site, and measure on a fixed schedule.
Is a caliper more accurate than a tape measure?
Not automatically. Per the break-even table, a caliper only beats a tape at half a centimetre of tape slop if you can repeat a pinch to 1.09 mm on a man and 0.80 mm on a woman. Below that quality the tape is the more repeatable instrument, and for most women the tape wins at every realistic pinch quality because the Navy equation's denominator is large.
How much does body fat have to change before I can be sure?
It depends on your own repeatability, which is what the calculator above is for. At the defaults: 1.58 points with a tape on one reading per site, 0.91 with three readings, and 4.16 with a smart scale carrying 1.5 points of noise. If you want to resolve one point, three readings per site with the tape or caliper will do it; eighteen will do it with the device.
How often should I measure?
Not more often than the arithmetic justifies. A change is visible once it exceeds 1.96·√2·σ, so at a genuine half point a month you need about three months with a tape at one reading per site, or a little under two at three readings. Measuring weekly against a three-month threshold only produces noise.
Why do my tape and my caliper disagree by five points?
Two reasons, and this page can only speak to the second. The equations carry their own population-fitted error, which this page does not quantify. The repeatability part is computed above: a lean man's tape carries 0.74 points of spread at half a centimetre of slop, so some of a five-point gap will be measurement rather than method. Our three-method page inverts the equations to show what disagreement looks like in centimetres and millimetres.
Should I buy a smart scale instead?
As a tracking device, understand what you are buying. A device contributing 1.5 points of spread can only resolve a 4.16-point change on single readings, and needs eighteen readings per comparison to resolve one point. It is convenient and it produces a daily number; that daily number is not a one-point measurement of anything. See our spec arithmetic page for what the printed figures are worth.
My body fat went up but my weight went down. Is my tape wrong?
Not necessarily. A tape estimate can move while weight falls if the waist falls more slowly than the rest of the body, and the BMI equation moves in lockstep with weight while the tape does not see weight at all. Check whether the change exceeds your threshold — 1.58 points at the defaults — before concluding either way.
Does it matter who takes the measurements?
Yes, and it is one of the largest effects this page can point at without measuring it. Two people place landmarks differently and pull a tape to different tensions, which is a systematic difference between sessions rather than a random one, so averaging will not remove it. If you can, be measured by the same person every time.
Is DEXA or hydrostatic weighing better?
This page has no numbers on them and will not produce any, because it has not measured against them. What can be said from the arithmetic here is structural: any method you use only occasionally gives you one reading with no repeatability estimate, so you cannot separate a real change from that method's own spread. A method you can repeat weekly answers a different and often more useful question.
Which method should a woman use?
On the numbers above, the tape, in four out of four computed female profiles. The Navy equation divides by the sum of waist, hip and neck, so the same half centimetre of slop costs a woman less than it costs a man, and the caliper's per-millimetre sensitivity runs slightly higher for women as well. If your tape work is sloppier than about a centimetre, the caliper takes over.
Related tools
body fat calculator · error budget for a single method · 3-site versus 7-site skinfold · the smallest change your tape can see · how to measure body fat at home · three methods on one body · Navy body fat calculator · what a scale's spec sheet is worth
Not medical advice. Every figure on this page is arithmetic this page performs by differentiating published equations and evaluating them on the numbers you enter. It measures repeatability only, not accuracy against any reference, and no instrument was tested for it. See our disclaimer.