Three systems, three sets of numbers
UK, US and EU sizing each use different increments and different reference bodies. Conversion charts exist, but they are approximations built on average figures, not equivalences.
A UK 8 is not a US 4. They are simply labels that a given brand has decided correspond.
Why they never standardised
Historical accident, mostly. The last comprehensive US anthropometric survey for apparel sizing was conducted in 1941 and does not reflect contemporary body shapes.
Brands have layered new products onto inherited charts ever since, and re-grading an entire catalogue is expensive enough that few attempt it.
The empirical picture
A 2025 survey of 687 shoppers across the US and UK found 91% said their size depends on the brand.
Research measuring over 1,000 garments across 50 UK brands found only 64% matched their own published chart within 1 cm, with 4.2 cm of average within-brand variance. 68% of garments labelled UK 12 fell outside industry-standard parameters.
What measurements change
Measurements remove the translation step entirely. Your waist is 78 cm. A brand charting a 30 inch waist is describing a body of about 76 cm, and you know your own number.
No system conversion required, and no ambiguity about which reference body is being used.
Size systems are a language for talking about bodies approximately. Measurements are the body itself. When you can use the latter, the former stops mattering.
The limits of measurement too
Measurements do not tell you how a garment drapes, and they do not account for ease, which is frequently unpublished. NMSU Extension specifies 3 to 4 inches of ease on the bust and 1 to 2 inches from nape to waist.
So measurements solve the language problem, not the garment problem.
What to do in practice
- Store measurements, not sizes.
- Compare against each brand's chart individually.
- Record the offset you find for each brand.
- Stop converting between systems, because the conversions are the unreliable part.
Further reading in using measurements across brands and the mismatch data.