Making a scanned PDF smaller without making it unreadable
A scanned PDF is large because it is a photograph of paper, and photographs are expensive. The size comes from three settings: how many pixels the scanner captured per inch, how many bits it used per pixel, and how it encoded the result. Changing those settings - not "running a compressor" - is what actually reduces size. The levers are worth applying in a fixed order, because the first one buys more than the next three combined and the last one can destroy the document.
Why scans are big: the arithmetic
A4 is 8.27 × 11.69 inches. At 300 dpi the scanner captures about 2481 × 3508 pixels, which is 8.7 megapixels per page. Stored as 8-bit greyscale that is roughly 8.7 MB of raw pixels per page before any compression. At 600 dpi the same page is 4962 × 7016 pixels, four times as many, and 34.8 MB before compression. A 40-page colour scan at 600 dpi can be in the hundreds of megabytes with no OCR layer at all.
| A4 page | Pixels | Raw, 8-bit grey | Typical encoded |
|---|---|---|---|
| 150 dpi, bitonal | 1241 × 1754 | 0.26 MB at 1 bit | 40–90 KB (CCITT G4) |
| 300 dpi, bitonal | 2481 × 3508 | 1.04 MB at 1 bit | 50–120 KB (CCITT G4) |
| 300 dpi, greyscale | 2481 × 3508 | 8.7 MB | 350 KB–1.2 MB (JPEG) |
| 300 dpi, colour | 2481 × 3508 | 26 MB | 1–3 MB (JPEG) |
| 600 dpi, colour | 4962 × 7016 | 104 MB | 4–12 MB (JPEG) |
The numbers are approximate and depend on content - a dense page of small text encodes worse than a page with a paragraph and a lot of white space. But the shape is reliable: bitonal text pages are two orders of magnitude smaller than colour photographs of the same page.
Lever one: resolution
Resolution is the setting with the most effect and the least subtlety. For text that will be read on screen or printed at A4, 300 dpi is the practical ceiling; 200 dpi is usually enough for a document that will only be read on screen; 150 dpi is legible for plain text and noticeably soft for small print or handwriting.
Going below 200 dpi is where documents start to fail: OCR accuracy drops on small type, signatures lose their detail, and a fax received at 200 dpi re-encoded at 100 dpi becomes a grey smear. If the page contains anything a human needs to compare - a stamp, a signature, a serial number - 300 dpi in bitonal or greyscale is the safe target.
Lever two: colour depth
Most office documents are black text on white paper. Scanning them in 24-bit colour stores three bytes for every pixel to describe something that needs one. Switching the scanner to greyscale or to black-and-white reduces the data by 3× or 24× before encoding even starts.
- Black-and-white (bitonal, 1 bit): best for typed text, contracts, invoices, photocopies. Edges stay crisp, files get tiny. Bad for photographs, pencil, colour stamps and anything with shading.
- Greyscale (8 bit): the default compromise. Handles pencil, faint print and stamps, and encodes well as JPEG.
- Colour (24 bit): only when colour carries information - highlighted clauses, a company letterhead used as evidence, an ID card, a product photo.
Lever three: encoding
A PDF stores each scanned page as an image inside the file, using whatever filter the producing software chose. Two families matter:
Lossless filters (Flate, CCITT Group 4) store exactly the pixels captured. Flate on a colour scan is roughly a 2:1 to 4:1 reduction - honest, but not enough to rescue a 100 MB file. CCITT Group 4 is designed for bitonal images and routinely achieves 20:1 or better on text, which is why 300 dpi bitonal pages end up under 120 KB.
Lossy filters (JPEG, and in modern scanners JPEG 2000) throw away detail the eye is unlikely to miss. At quality 75 a greyscale text page compresses to a few hundred kilobytes with no visible damage at reading size. At quality 30 the ringing artefacts around letters are permanent, and each re-encode makes them worse.
Every lossy re-encode is destructive. If you must compress twice, always compress from the original scan, never from the already-compressed PDF.
Lever four: the pages you do not need
Before touching quality, remove what should not be there. Blank backs from duplex scanning, separator sheets, duplicate pages from a double-feed, and cover sheets that were scanned twice routinely account for 10–30 % of a batch. On a 60-page scan of 40 MB, dropping 15 blank pages is 10 MB saved with no quality cost at all, and it makes the document easier to read.
This is the one lever PDFNest can help with directly: the split tool can drop pages by keeping only the ranges you want, and the merge tool can rebuild the batch in order afterwards. Neither re-encodes anything, so the pages that remain are untouched.
What does not work
- "Compress PDF" on a born-digital file. If the PDF was exported from a word processor, its text is already compact; the images inside it are the only target, and re-encoding them at low quality makes a clean document worse for a small gain.
- Removing the text layer to save space. The OCR layer costs tens of kilobytes and is the only thing that makes the document searchable. Deleting it saves almost nothing and makes the file much less useful.
- Printing to a virtual PDF printer at low quality. This re-rasterises the whole page - including any vector content - and often produces a larger file than the original scan.
- Compressing a signed or archival document. Re-encoding changes the bytes, breaks digital signatures, and may fall foul of a records policy that requires the original capture to be preserved.
Target sizes and what they imply
| Destination | Practical limit | Settings that fit |
|---|---|---|
| Email attachment | 10–20 MB total | 200–300 dpi greyscale, JPEG 70 |
| Government or bank portal | 2–10 MB per file | 300 dpi bitonal for text pages, or split into parts |
| Court filing | Often 10–35 MB, format rules apply | 300 dpi bitonal plus OCR, follow the court's spec |
| Internal archive | No practical limit | Keep the original capture, 300–400 dpi, lossless |
When a portal limit is the problem, splitting is often better than degrading the image: a 12 MB scan split into two 6 MB parts is accepted everywhere and loses nothing. Use the split tool with page ranges at a sensible boundary, and name the parts so they sort correctly.
Checking the result before you send it
Open the new file at 100 % zoom - not fit-to-window, which hides everything - and read the smallest text on the densest page. Check a signature and a date stamp. If the document has an OCR layer, search for a word you know is on page 3; if the search fails, the text layer was damaged by the process. Then check the page count against the original: a drastic size drop usually means pages were lost, not that compression worked.
Frequently asked
Can PDFNest compress the images inside a PDF?
No, and the site does not pretend to. Re-encoding images would mean either shipping a codec into the page or quietly destroying detail. What it can do is remove pages you do not need, split the file into parts that fit a limit, and rebuild the order afterwards - all without touching the pixels.
Why did my scan get bigger after I edited it?
Many editors rewrite the file on save: the original image may be re-encoded, the cross-reference table rebuilt, or an OCR layer regenerated. Always keep the original scan and check the size of the exported copy before replacing anything.
Is 150 dpi enough for a contract?
For reading on screen, usually yes. For anything that may be printed, disputed or read by OCR years later, 300 dpi bitonal is a better default - it is barely larger than 150 dpi bitonal and much more robust.
Related reading: password-protected PDFs and naming archived bundles.
Written by the PDFNest team. Last updated November 19, 2026.