Is your PDF scanned?
Open it and try to select a word. If you can highlight and copy text, it is a normal (digital) PDF: choose Normal PDF, which is faster and more accurate. If nothing selects, or the whole page highlights as one image, it is scanned and needs Scanned (OCR).
Typical scanned PDFs:
- Paper documents scanned on a printer or document scanner
- Phone photos of receipts, invoices or forms saved as PDF
- Fax archives and old records digitised as images
- PDFs some banks and government services generate as images
How to convert a scanned PDF to Excel
- Drop the PDF on the converter on this page.
- Keep Tables only and choose Scanned (OCR).
- Choose Accurate for small print, dense tables or poor scans. Fast is fine for clear, large text.
- Convert and download the
.xlsxfile. The first run downloads the OCR language data (a few MB), which your browser then caches. - Check the numbers against the original before using them.
How it works
- OCR. Each page is rendered as an image and Tesseract, an open-source OCR engine, finds every word and its position.
- Table reconstruction. Words are grouped into rows by vertical position and into columns by the gaps between them, the same method used for digital PDFs.
- Cleanup. Repeating headers and footers are removed, wrapped cells are merged, and numbers, currency amounts and dates are converted to real Excel values.
The workbook also includes a confidence sheet that shows how sure the OCR engine was about each cell, so you can find the cells most likely to need checking.
Getting a good result from a scan
OCR quality depends mostly on the image. In order of impact:
- Resolution. Scan at 300 DPI. Below 200 DPI, small digits are often misread. 400–600 DPI helps with very small print but makes the file larger.
- Straight pages. A tilted page makes rows drift across columns. Use your scanner's auto-straighten (deskew) option, or re-scan.
- Contrast. Black text on a white background works best. Faded thermal receipts and coloured backgrounds lower accuracy.
- No shadows or folds. For phone photos, shoot flat, from directly above, in even light. Use a scanning app that crops and flattens the page.
- One page per image. Two pages photographed side by side confuse column detection.
Limitations to know about
- OCR currently reads English text. Other Latin-alphabet languages may partly work, but accents and special characters are often misread.
- Handwriting is not reliably recognised. OCR is designed for printed text.
- Check figures where 1 and 7, 5 and 6, or 0 and 8 could be confused, especially on low-quality scans.
- Very long scanned documents take time, because OCR runs on your own computer rather than a server.