How Accurate Is OCR? What Decides Whether It Works
OCR is highly accurate — often above 98% — on clean, high-resolution images of printed text in a supported language, but accuracy falls sharply with low resolution, poor contrast, skew, unusual fonts, or handwriting. “How accurate is OCR” has no single answer because the number is decided almost entirely by the quality of the input, not the tool. Here’s what moves it, and how to give OCR the best chance.
This guide explains the five factors that decide OCR accuracy and how to improve each, so you get the most out of the image to text (OCR) tool.
What “accuracy” even means
OCR accuracy is usually quoted at the character level — the percentage of characters read correctly. That sounds reassuring until you do the maths: 98% character accuracy still means roughly one wrong character every two lines, so a page can need a few corrections even when the tool is performing well. Word-level accuracy is lower again, because a single wrong character spoils the whole word. Treat OCR as getting you a fast, mostly-right draft to proofread — not a guaranteed perfect transcription.
The five things that decide the result
- Resolution. The single biggest factor. Around 300 DPI is the sweet spot for printed text; below about 200 DPI, letters blur together and accuracy drops fast. A photo taken close and in focus beats a distant, low-resolution one.
- Contrast. Crisp black text on a white background reads best. Faint print, coloured backgrounds, watermarks and shadows all cost accuracy.
- Skew and distortion. Text should be level. A tilted scan, a curved page near a book’s spine, or perspective from a phone photo all confuse line detection.
- Font and layout. Standard serif and sans-serif fonts are recognised best. Decorative fonts, very small type, tight kerning, and multi-column layouts are harder. Tables and forms add structure the tool has to untangle.
- Language and script. Accuracy depends on the OCR being run with the right language model; accented characters and non-Latin scripts need explicit support.
Where OCR struggles most
- Handwriting — a different problem from printed OCR, and far less reliable, especially cursive.
- Low-quality scans and faxes — compression artefacts and speckle mimic characters.
- Screenshots of tiny text — too few pixels per letter.
- Stylised or condensed fonts — logos, receipts, dot-matrix print.
How to get a better result
- Start with the highest-resolution image you can — rescan at 300 DPI rather than upscaling a small one (upscaling adds no real detail).
- Increase contrast and straighten the image before running it — even a quick crop and auto-contrast helps.
- Run it, then proofread — check the tricky pairs OCR confuses:
0/O,1/l/I,rn/m,5/S. - Prefer the digital original. If the text exists as a real PDF anywhere, extract it with the PDF text extractor instead — that’s 100% accurate because no recognition is involved.
Drop your image or scan into the image to text (OCR) tool — it runs entirely in your browser, so the image is never uploaded. For when to use OCR versus plain extraction, see scanned vs searchable PDF.
Frequently asked questions
How accurate is OCR?
On clean, high-resolution printed text it’s often above 98% character accuracy; on low-resolution scans, unusual fonts or handwriting it can be far lower. The input quality decides the result more than the tool does.
What resolution do I need for good OCR?
Around 300 DPI for printed text. Below roughly 200 DPI, characters blur and accuracy falls quickly.
Can OCR read handwriting?
Not reliably. Handwriting recognition is a harder, separate problem, and cursive especially is error-prone.
Why does OCR confuse certain characters?
Similar shapes trip it up — 0 and O, 1, l and I, rn and m. Proofreading these pairs catches most errors.
Is it better to OCR a scan or extract from a digital PDF?
Always prefer the digital PDF if it exists — extracting real text is 100% accurate, whereas OCR is a best-effort recognition of an image.
Related reading
Last updated: 16 August 2026.