QR Codes, Explained

After reading this you will understand how a QR code packs text into a grid of black and white squares, how error correction lets a scratched code still scan, and how to pick a size and correction level that actually works in print.

A QR code is a two-dimensional barcode: a square grid of black and white cells that encodes text. Point a camera at it and you get back a string, usually a URL like https://example.com. The grid you see is not decoration. Every cell is a bit, and the pattern includes finder squares, timing lines, format bits and redundant recovery data.

Here is the hook. Take any working QR code and cover up to 30 percent of it with a logo or a coffee stain. If it was made at error-correction level H, it still scans. That resilience is built into the math, and this article shows you where it comes from.

What a QR code actually stores

A QR code stores a byte string plus enough extra bytes to repair errors. The visible grid is organized into fixed patterns and data:

Finder patterns

The three large squares in the corners. A scanner uses them to locate and orient the code.

Timing patterns

Alternating black and white lines between the finders that fix the module coordinate grid.

Format and version information

Small reserved regions that tell the decoder the error-correction level and the mask pattern used.

Data and error-correction codewords

The rest of the grid, holding your payload and the Reed-Solomon recovery bytes.

The size of the grid is set by the version. Version 1 is 21 by 21 modules. Each version up adds 4 modules per side, so version 2 is 25 by 25, version 10 is 57 by 57, and version 40 is 177 by 177. The library picks the smallest version that fits your data at the chosen error-correction level.

When to use one, and when not

QR codes are worth it when a person will point a phone at a physical or on-screen thing. Restaurant menus, event tickets, Wi-Fi join codes, contact cards on a business card, and payment links are all good fits. The reader gets a long or awkward string without typing it.

They are a poor fit when the destination is already easy to type or when the code will be tiny and low contrast. A 12-character coupon code printed as plain text beats a QR code that a phone struggles to focus on. For linear product identifiers such as EAN or UPC, use a one-dimensional barcode instead; see the Barcode Generator.

The capacity and error-correction math

The total number of codewords in a version is split between data and error correction. If a version has n total codewords and reserves 2t of them for correction, Reed-Solomon can repair up to t corrupted codewords:

Here n is the total codewords in the block, k is the number of data codewords you supply, and t is how many wrong codewords can be fixed. Each codeword is one byte. The four error-correction levels set the ratio of recovery bytes:

The trade is direct. At the same version, moving from L to H replaces data bytes with recovery bytes, so the same message needs a larger, denser grid. Level M is the everyday default because 15 percent recovery survives normal print wear without bloating the code.

A worked example with the demo data

Reading and decoding a code

Decoding reverses the process. A reader locates the three finder squares, corrects for perspective and rotation, samples each module as black or white, reads the format bits to learn the level and mask, un-masks the grid, then runs Reed-Solomon to repair any errors before extracting the byte string.

The mask step matters more than it looks. Before finalizing, the encoder XORs the data region with one of eight mask patterns and keeps the mask that avoids large solid blocks and false finder-like shapes. This keeps the code readable and helps the camera. When you decode an image on this page, the decoder reads the stored mask index and reverses it automatically. Everything runs locally in your browser, so the uploaded image never leaves your machine.

Common mistakes

Inverting the colors. Most scanners expect dark modules on a light background. A light-on-dark code sometimes works and sometimes does not, depending on the reader. If contrast is low, decoding fails. Keep a luminance difference of at least 40 percent between foreground and background.

Cropping the quiet zone. Designers often trim the white margin to fit a layout. The 4-module border is part of the specification. Without it, the scanner cannot separate the finder patterns from surrounding artwork.

Over-encoding. Putting a 400-character message in a code forces a high version with tiny modules. Shorten the payload. Link to a page rather than embedding a whole vCard when the code will be printed small.

Logo too large. A center logo works only because error correction repairs the covered modules. At level M you have roughly 15 percent to spend. Cover more and the code stops scanning. Raise to level Q or H before adding a big logo, and re-test on a real phone.

Related tools

For linear retail codes, the Barcode Generator handles EAN, UPC and Code 128. If you need filler graphics while building a layout around a code, the Placeholder Image Generator makes sized boxes with labels. To pack many small images into one file, see the Sprite Sheet Generator & Splitter. For stylized black-and-white output from photos, the Dithering Tool and the ASCII Art from Image tool are both here.

Frequently asked questions

Related tools

Handwriting Font Maker

Turn your handwriting into a real font: print the template, write, photograph, download a TTF.

Barcode Generator

EAN, UPC, Code 128 and more as PNG or SVG.

ASCII Art from Image

Turn any picture into text-mode art.

Dithering Tool

Floyd-Steinberg, Atkinson and ordered dithering effects.

Placeholder Image Generator

Custom-size placeholder images with labels.

Sprite Sheet Generator & Splitter

Pack images into a sprite sheet or slice one apart.