QR Code Error Correction Levels: What L, M, Q, and H Actually Do
QR code error correction is the built-in redundancy that lets a code keep scanning after part of it is scratched, smudged, covered, or torn away. The QR standard, ISO/IEC 18004, defines four levels: L survives roughly 7 percent damage, M 15 percent, Q 25 percent, and H 30 percent.
That one dropdown in your generator decides more real outcomes than any styling choice. It sets how much data fits in a grid, how small you can print, whether a centre logo is safe, and whether a scuffed label still reads after a year on a job site. Here is what the levels actually do, what they cost, and which one to pick.
How QR error correction actually works
Every QR code stores its content as 8-bit chunks called codewords. When a generator builds your code, it feeds those codewords through Reed-Solomon math, the same error correction family used on CDs, DVDs, and deep space radio links, and appends extra check codewords calculated from your data.
The check codewords are not copies. They work more like a solvable equation: if some codewords are lost because modules got scratched out or misread, the scanner plugs the surviving values into that equation and rebuilds the missing ones exactly. Within the level's limit, the decoded result is bit for bit identical to the original, never a guess.
Two details matter in practice. The percentages apply to codewords, not surface area, so a 30 percent budget does not guarantee survival of every possible 30 percent hole. And in larger codes the data is interleaved across several correction blocks, which spreads one localized scratch thinly across many blocks instead of letting it destroy a single block outright.
You never tell the phone which level you used. The choice is stored in the format information strip next to the three corner squares, so every scanner detects it automatically.
The four levels at a glance
| Level | Recoverable codewords | Bytes at Version 5 (37x37) | Capacity vs L | Best for |
|---|---|---|---|---|
| L | ~7% | 106 | baseline | screens, maximum data |
| M | ~15% | 84 | 21% less | everyday links, the default |
| Q | ~25% | 60 | 43% less | print, packaging, outdoors |
| H | ~30% | 44 | 58% less | logos, harsh handling |
The capacities come straight from the ISO/IEC 18004 capacity tables for a Version 5 code in byte mode. The ratios hold roughly across all 40 versions, from the smallest 21x21 grid up to the 177x177 giant that stores 2,953 bytes at L but only 1,273 at H.
What the higher levels really cost
Redundancy is paid in one of two currencies. Hold the grid size fixed and you store fewer characters. Hold the data fixed and the grid grows. The second cost is the one that bites on physical products, because a bigger grid at the same print size means smaller modules.
| Level | Version for a 50 character URL | Grid | Module size at 20 mm print |
|---|---|---|---|
| L | Version 3 | 29x29 | 0.69 mm |
| M | Version 4 | 33x33 | 0.61 mm |
| Q | Version 5 | 37x37 | 0.54 mm |
| H | Version 6 | 41x41 | 0.49 mm |
Same URL, four different physical codes. Moving from L to H doubles the module count from 841 to 1,681 and shrinks each module by nearly 30 percent at a fixed 20 mm print. That interaction drives most of the advice below.
Which level should you use
L: screens only
A code shown in an app, on a slide, or on a TV cannot be scratched, so damage tolerance buys you nothing there. Pay only the 7 percent overhead and keep the modules big. L is also the right call when you are pushing capacity limits, like a long embedded contact card or plain text payload.
M: the default for a reason
Most generators default to M because 15 percent covers ordinary life: a thumb over one edge, light glare, a slightly soft print. If you have no special risk, M is correct and costs only about a fifth of your capacity.
Q: anything that gets printed
Paper wears. Packaging gets scuffed in transit, stickers pick up dirt, menus collect stains. Q's 25 percent budget absorbs that for around 43 percent of capacity, which rarely matters because printed codes usually carry short URLs anyway.
H: logos and hard lives
H exists for codes that will be deliberately covered or predictably damaged: logo overlays, garment tags that go through the wash, asset tags on tools, and anything living outdoors without lamination.
Why a logo overlay demands H
A logo in the middle of a QR code is deliberate, pre-spent damage. A typical centred logo covers 10 to 20 percent of the modules. That alone exceeds L's 7 percent budget, and at M it can consume the entire 15 percent, leaving zero margin for a fingerprint or a scratch on top. At H, a 15 percent logo still leaves about half the correction budget for real wear.
Two hard limits apply. Keep the logo under about 20 percent of the area even at H, and never let it touch the three big corner squares or the smaller alignment squares. Error correction can rebuild data modules, but the scanner needs those locator patterns intact just to find the grid, and no level can recover them.
Damage tolerance and print size pull against each other
Print size changes what a fixed physical injury costs. A 2 mm wide scrape across a 20 mm label wipes out about 10 percent of the symbol, which already stresses M. The same scrape across a 50 mm poster code costs 4 percent, which even L shrugs off. Bigger prints need less correction, smaller prints need more.
Here is the trap: small labels are exactly where raising the level hurts most, because the version jump shrinks modules that are already tiny. On a 20 mm label, the table above shows H pushing modules down to 0.49 mm, where cheap printing and everyday wear start eating module edges, so you trade scratch resistance for a soft, mushy print. Our guide to blurry QR codes and their fixes covers that failure mode in detail.
The fix is to shorten the data instead of lowering the level. A 28 character short link fits Version 3 (29x29) even at Q, the same grid the 50 character URL needed at L. If you are producing sticker sheets or product labels, build them with our QR code for labels generator and keep the payload short so you can afford Q or H without shrinking modules past the safe zone.
Real failures and what would have saved them
Scratched asset tags. Tags on power tools and IT equipment collect gouges from keys, drill bits, and shelving. A gouge that kills 12 percent of codewords is fatal at L, borderline at M, and routine at Q or H. If you label equipment, generate at H with our asset tag QR generator and laminate where you can.
Faded thermal labels. Direct thermal shipping labels fade after months of light and heat, and here error correction barely helps, because fading is global: every module loses contrast at once and the scanner cannot separate dark from light anywhere in the symbol. That failure needs contrast and reprint fixes, not a higher level; our QR code not scanning guide walks through the diagnosis order step by step.
Cracked garment labels. Printed care labels crease, abrade, and survive 30 plus wash cycles. Heat pressed codes crack along fold lines, a classic localized damage pattern that H handles well. If you tag clothing you probably need both symbologies: retail checkout runs on barcodes, which our garment barcode generator produces to spec, while any QR carrying care or authenticity info should always be generated at H.
The short version
L on screens, M for everyday links, Q for anything printed, H for logos and rough handling, and shorten the URL before you ever lower the level. For generators, decoders, and a safe scanner in one place, browse our QR tools, or start with the rest of our guides if you want the background before you build.