# Format information — the 15 bits everything depends on

> Format information is 15 bits: 2 for the error-correction level, 3 for the mask pattern, and 10 BCH(15,5) check bits, all XORed with the fixed mask 101010000010010. It is written twice beside the finder patterns, because a decoder can read nothing else until it has decoded these bits.

Source: https://useqr.app/docs/spec/format-information · Last reviewed 2026-08-21 · UseQR is free forever, MIT licensed, no signup.

---

## What the 15 bits say

[Format information](/glossary/format-information) carries the two settings a decoder must
know before touching the data region:

| Bits | Content |
|---|---|
| 2 | [Error-correction level](/glossary/error-correction-level): **L = 01, M = 00, Q = 11, H = 10** |
| 3 | [Mask pattern](/glossary/mask-pattern) number, 000–111 |
| 10 | BCH(15,5) error-correction bits over the 5 data bits |

Note the level encoding is not in strength order — M is `00`, below L's `01`. It is a spec
quirk worth knowing when reading raw bits.

The full 15-bit word is then **XORed with the fixed mask `101010000010010`**, which
guarantees the format region is never all light (an all-zero format would otherwise render
as blank background and be undetectable).

Worked example — level M, mask 5: data bits `00101`, BCH-coded to `001010011011100`,
XORed to **`100000011001110`**. Those are the exact modules written into the symbol.

## Where it lives

The word is written **twice**:

- Copy 1 wraps the top-left finder pattern, along row 8 and column 8.
- Copy 2 is split: part under the top-right finder, part beside the bottom-left one.

One extra module — the **dark module** at row 4 × version + 9, column 8 — sits permanently
dark next to the second copy. It carries no information; it is a fixed feature of
[the symbol's anatomy](/docs/spec/qr-code-anatomy).

## Why it is the most protected data in the symbol

Every other bit in a QR code is recoverable through Reed–Solomon — *if* the decoder knows
which mask to remove and which block structure to expect. Both facts live in the format
information. Corrupt them and the decoder unmasks with the wrong pattern and
de-interleaves with the wrong geometry: the payload becomes unrecoverable even when every
data module is intact. So the standard defends these 15 bits three ways:

1. **BCH(15,5)** — 10 check bits for 5 data bits, able to correct up to 3 bit errors per
   copy. That is a far higher redundancy ratio than the data region gets at any level.
2. **Two complete copies** in different parts of the symbol, so localised damage rarely
   takes both.
3. **A brute-forceable space.** There are only 32 valid format words; a desperate decoder
   can compare a damaged read against all of them and pick the nearest — which is exactly
   what most implementations do.

## What this means for styling

The format regions hug the finder patterns — precisely where
[frames and labels](/docs/design/qr-code-frames-and-cta-labels) and oversized custom eyes
tend to encroach. A styled code that repaints row 8 or column 8, or lets a logo drift
toward a corner, is spending its most expensive bits first. Damage simulations bear this
out: flipping a handful of modules beside two finders can kill a code that would have
survived 25% damage in the data region. Keep effects away from the finder surrounds, and
[verify the rendered code decodes](/validate) after any styling.

## FAQ

### What is format information in a QR code?
Fifteen bits beside the finder patterns recording the error-correction level (2 bits) and mask pattern (3 bits), protected by 10 BCH check bits and XORed with the fixed mask 101010000010010. It is stored twice for redundancy.

### Why is format information stored twice?
Because nothing else in the symbol can be decoded without it. Two copies in different locations mean localised damage — a scratch near one finder — rarely destroys both, and the BCH code can repair up to 3 bit errors in each copy.

### What is the single black square that is always in the same place?
The dark module, at row 4×version+9, column 8, beside the bottom-left finder. It is always dark, carries no data, and is simply a fixed structural feature defined by the standard.

### What happens if format bits are damaged?
The decoder tries BCH correction (up to 3 bit errors), falls back to the second copy, and as a last resort compares against all 32 valid format words. If all of that fails, the code is unreadable even when the data region is perfect.

## Try it

- https://useqr.app/url
- https://useqr.app/validate
- https://useqr.app/scan