Methodology
Card-calibrated, a measurement here is good to about ±0.15 mm of inside diameter — about a fifth of a US size. This page shows where that number comes from, what it does not cover, and when to stop trusting a screen and see a jeweller.
Step one
A ring size is not a physical quantity. The physical quantity is the inside circumference of the band, in millimetres — or equivalently its inside diameter, since one is π times the other. Every national system (US, UK, EU/ISO, Indian, Japanese) is a naming scheme laid over that single measurement, which is why the ring size chartcan convert between all five without any of them disagreeing.
Sizes here derive from the ISO 8653 relationship circumference (mm) = 36.537 + 2.5535 × US size. One whole US size is therefore 2.5535 mm of circumference, or 0.813 mm of diameter. That last figure is what makes a tolerance meaningful: ±0.15 mm of diameter is ±0.18 of a US size, ±0.3 mm is ±0.37, and ±0.5 mm is ±0.62.
Nothing here is estimated from a photograph, a hand model or a machine-learning guess. The number you get out is the number you put in, converted.
Step two
A CSS pixel has no fixed physical size. Browsers fix it at 1/96 of an inch only for printing; on a display it is whatever the panel density and the device pixel ratio make it. The same 100-pixel circle is about 26 mm on an older desktop monitor and about 16 mm on a dense phone. An on-screen sizer that draws a circle without calibrating first is drawing a guess, and the guess can be wrong by 40%.
Calibration produces one number, cssPxPerMm, and every circle drawn afterwards derives from it. Card calibration gets that number directly:
cssPxPerMm = card width on screen (px) ÷ 85.6
85.6 mm by 53.98 mm is the ID-1 format defined by ISO/IEC 7810. Bank cards, credit cards and most national ID cards and driving licences are manufactured to it worldwide, to a tolerance of a few hundredths of a millimetre — tighter than you can see. That makes the card in your pocket a better reference object than most household rulers.
The other two methods work backwards from pixel density instead: cssPxPerMm = PPI ÷ devicePixelRatio ÷ 25.4, with the PPI either looked up from a device table or computed from your native resolution and screen diagonal. A calibration implying a density below 72 or above 700 PPI is rejected outright, because no real display sits outside that range.
Step three
Four things can go wrong in a card-calibrated measurement. Each is small, and they are independent, so they combine in quadrature rather than simply adding. The figures below are for a 17.3 mm ring, about a US 7.
| Source | Contribution | Why |
|---|---|---|
| Card edge alignment | ±0.05 mm | You line the on-screen rectangle up against a real card by eye. An error of one CSS pixel on a rectangle roughly 320 px wide is a 0.3% scale error, which is 0.05 mm on a 17.3 mm ring. |
| Slider quantisation | ±0.03 mm | The calibration slider moves in half-pixel steps, so the scale itself is quantised to about 0.15%. On a 17.3 mm ring that is 0.03 mm. |
| Matching the ring to the circle | ±0.10 mm | The size slider steps in 0.05 mm. Judging when the drawn circle meets the inside edge of a band is worth about two of those steps either way. |
| The ring not lying flat | ±0.05 mm | A band with a domed profile, or one resting on a raised stone, sits slightly off the glass and reads a little wide. |
| Combined | ±0.13 mm | Root sum of squares. The site publishes ±0.15 mm rather than ±0.13 mm, because rounding an error bar down is how error bars stop meaning anything. |
In US sizes, ±0.13 mm is ±0.16 of a size. That is enough to tell a 7 from a 7¼, and not enough to argue about anything finer.
The three methods
±0.18 US size
Compares the screen against a physical object whose size is fixed by international standard. It assumes nothing at all about your hardware, which is why it is the recommended method.
Breaks when: A worn or bent card, or one held at an angle to the screen. Cards are rigid by standard, so a visibly curved card should not be used as the reference.
±0.37 US size
Your logical resolution and device pixel ratio are matched against a table of phones and tablets with published pixel densities. The stated figure is deliberately conservative: it covers the case where two different panels share one signature and their real densities differ by up to about 1.5%.
Breaks when: Unknown hardware. Desktops are absent from the table entirely, because a browser window size says nothing about the monitor — auto-detect declines rather than guessing.
±0.62 US size
Pixel density is derived from your native resolution and the diagonal you type in. Marketing diagonals are rounded — 13.3 quoted for a 13.6-inch panel is a 2% scale error — and the stated figure covers rounding of that order.
Breaks when: A diagonal that is simply wrong. Entering 15 for a 15.6-inch laptop is a 4% error, which no stated tolerance covers. Check the spec sheet, or use a card instead.
Every result the calculator shows carries the tolerance of the method that produced it, in millimetres and in US sizes. And if you would rather check the calibration than take our word for it, the sizer draws a 85.6 mm bar once you are calibrated — hold your card against the glass. If they match, everything downstream is sound.
The larger errors
Calibration error is the smallest problem in ring sizing. Everything below moves the answer further than ±0.15 mm ever will, and no measuring method removes them.
Two practical consequences. Measure in the late afternoon at room temperature, so you size at your largest and the ring fits all day rather than part of it. And if you fall between two sizes, go up: a loose ring can be resized or beaded, while one that will not pass the knuckle cannot be worn at all. The guide to measuring ring size at home covers each method in order of accuracy.
Limits
A jeweller’s steel mandrel and set of test rings remain the reference standard, and being measured costs nothing. This tool exists to answer the question well when that is not practical — not to claim it has replaced it.
Corrections
Every chart on the site is generated from the single ISO 8653 formula above rather than typed in, so no two tables can drift apart. The UK letter, Indian and Japanese columns are the exception — those systems follow no clean formula and are looked up from published half-step tables.
The device density table is extended as new hardware appears, and auto-detect declines on anything it does not recognise rather than guessing. If you can show that a size, a density or a tolerance here is wrong, please tell us — corrections to the charts are answered first.