How We Calculate Shoe Sizes
We don't pretend a single "true" size exists — inside one US 9, real shoes vary by about 26mm between models. Instead we combine deterministic conversion formulas with measured brand-and-model fit data, then publish a best fit plus a range and a confidence score, showing exactly where every number comes from. That transparency is why we land the right size more often than a one-size-fits-all chart.
1. Why brand sizes don't match
There is no mandatory sizing standard — ISO 9407 is voluntary — so every brand designs around its own lasts (the foot-shaped moulds shoes are built on), its own design philosophy, and its own grading scheme. The result is that the same label means different physical lengths.
Measured internal length inside a single US 9 ranges from 255 to 281mm across brands — a spread of about 26mm, which is more than two full size steps. The variation is not just between brands: across Adidas models alone, US 9 measures 260.1–280.4mm. That is why accuracy is only achievable at the model and last level, never at the brand level — and why a universal conversion chart cannot be exact.
2. Our two-layer data model
Layer A — deterministic formulas
Every international system is formally defined and converts deterministically through one canonical axis: foot length in millimetres. These formulas are fully reproducible, and we show them right next to each number.
| System | Formula | Note |
|---|---|---|
| US men's | size = (foot_length_in × 3) − 22 | Barleycorn step (1/3 inch). |
| US women's | size = US men's + 1.5 | Typical brand offset between genders. |
| UK | size = (foot_length_in × 3) − 23 | Same step as US, different origin (UK = US − 1). |
| EU | size = foot_length_cm × 1.5 (rounded) | Paris point ≠ barleycorn — EU never maps exactly. |
| CM (Mondopoint) | size = foot_length_mm / 10 | The canonical axis. Every other system is generated from it. |
EU sizing is always rounded — the Paris point (≈6.67mm) and the barleycorn (≈8.47mm) use different steps, so an exact EU equivalent does not exist. We round to the nearest whole size and label it as the closest honest match rather than pretending it is precise.
Layer B — empirical brand fit
On top of the formulas we add measured last and insole lengths for specific
brands and models, plus a fit_offset — for example, a brand
that consistently runs half a size small carries a +0.5 offset.
Layer B is where the real value lives: it answers "does this shoe run small?"
in a way a pure formula never can.
3. How we score confidence
Every recommendation carries a confidence score from 0 to 1. It is an honest, reproducible function — never a marketing number — built as:
confidence = (base + lab_bonus + sample_size_bonus) × spread_penalty
| Component | Value | Why |
|---|---|---|
| Base — 1 source | 0.30 | A single unverified data point. |
| Base — 2 sources | 0.50 | Two independent sources agree. |
| Base — 3+ sources | 0.65 | Triangulated across three or more. |
| Lab measurement bonus | +0.15 | A physical lab measurement anchors the number. |
| Sample-size bonus | +log₁₀(N) / 4 (max +0.20) | More measurements add certainty, with diminishing returns. |
| Spread penalty — over half a size | × 0.7 (> 4.23mm) | Sources disagree by more than half a size. |
| Spread penalty — over a full size | × 0.5 (> 8.47mm) | Sources fundamentally disagree about foot length. |
We publish a recommendation as confident only when its score reaches our threshold of 0.70 or higher. Below that, we keep the number but flag the uncertainty openly rather than hiding it.
4. Where our data comes from
We triangulate each number across independent sources, in this order of trust. Crucially, we publish our own aggregates and calculations with attribution — we never republish anyone else's tables.
| Priority | Source | Role |
|---|---|---|
| 1 | Our own fit feedback ("how did it fit?") | First-party data, collected from day one — our growing moat. |
| 2 | Public lab measurements (RunRepeat, Solereview) | Calibration anchors only — we cite them, we never copy their tables. |
| 3 | Official brand charts | One source among many, always stored with a version date. |
| 4 | Retailer review aggregates | Directional crowd signal with sample size, window/scope notes where available, and lower weight than lab or first-party data. |
Review aggregates are never treated as a direct measurement. They are a directional signal: useful when many buyers report the same fit pattern, weak when the source lacks a sample window, clear model scope, or repeatable collection notes. When an aggregate disagrees with a brand chart or lab measurement, the spread lowers confidence instead of being averaged away.
5. Best fit vs strict conversion
We always show both answers, and we keep them clearly separate.
Strict conversion (Layer A) is the formula result —
for example US 10 (men) = UK 9 = EU 44 = 28.0cm — shown with
its formula and a note that EU rounding is unavoidable.
Best fit (Layer B) is the recommendation we'd actually bet on: a size, a likely range, a confidence score and its provenance. For example:
Best fit: US 9.5 (likely 9–10) · confidence 0.82 · based on 14 measurements and 320 fit reviews.
Because roughly 1mm is already a perceptible change in fit, we give a range instead of a falsely precise single number — and we never present a single figure as an absolute without that range and confidence.
6. How we keep numbers current
- We review top-brand records monthly and re-check broader official chart sources at least quarterly, because brands can change lasts between seasons.
- Every number on the site carries a "last verified" date, so that per-record date is the source of truth for freshness.
- When our sources disagree by more than half a size, we mark the record low-confidence and show you the disagreement instead of hiding it.