Why PDF files get so large, and what actually shrinks them

Guides · Updated 12 August 2026 · 8 min read

A ten-page text document should be well under a megabyte. When it is forty, one specific thing has almost always happened — and knowing which one tells you exactly what to do about it, rather than running the file through a compressor and hoping.

Find the weight before trying to remove it

Compression tools apply the same treatment to every file. It is far more effective to work out where the bytes are first, because the answer is usually a single dominant cause:

• Divide the size by the page count. Under 100 KB per page is a normal text document. 300 KB to 1 MB per page suggests images at reasonable resolution. Several MB per page means a scan, or photographs at full camera resolution.
• Ask whether the pages contain photographs. If yes, images are the cause; it is almost never anything else.
• If the pages are plain text and the file is still huge, suspect embedded fonts, an accumulated revision history, or an invisible full-page image behind the text.
• Try selecting the text. If you cannot, the page is a picture and you are dealing with a scan, whatever it looks like.

Images: where almost all the weight is

Two behaviours account for most oversized PDFs.

Resolution is stored, not displayed size. A PDF keeps the image at the resolution it was given. Placing a 6000 × 4000 pixel photo from a phone into a document and scaling it to a 6 cm square changes only the drawing matrix — the 24-megapixel image is still in the file. Ten such photos produce a 40 MB document with ten thumbnails in it.

Scanners default too high. Many scanners default to 600 dpi in colour, which is appropriate for archiving artwork and wildly excessive for a signed form. An A4 page at 600 dpi in 24-bit colour is roughly 100 megabytes uncompressed; even after JPEG compression it lands in the several-megabytes range. The same page at 200 dpi greyscale is perhaps a twentieth of that and looks identical on screen.

The useful reference points: 96–150 dpi for documents that will only be read on screen, 200–300 dpi for anything that will be printed or OCR'd, and above 300 dpi only for archival reproduction. Going beyond 300 dpi does not improve OCR accuracy — it is a common and expensive misconception.

Colour mode matters as much as resolution. A black-and-white document scanned in colour carries three channels of noise where one would do. Scanning text in greyscale, or in bitonal mode with a fax-style compressor, routinely cuts the file by an order of magnitude.

Fonts, revisions and other quieter causes

When images are not the answer, the usual suspects are:

• Fully embedded fonts. A PDF should embed only a subset — the glyphs the document actually uses. Some exporters embed the entire typeface, and a full CJK font can run to several megabytes on its own. A document using four weights of two families can carry tens of megabytes of typeface for a few pages of text.
• Accumulated revisions. Because PDF writers can append changes rather than rewrite the file, a document edited and re-saved many times can contain all its previous states. The file grows every time and none of the growth is visible.
• Duplicated resources. Merging documents that share a logo can embed that logo once per source file. Good writers detect and share identical objects; not all do.
• Vector artwork with too many points. A map or CAD export can hold hundreds of thousands of path segments. It compresses poorly because it is not repetitive, and there is no resampling to fall back on — it has to be simplified at the source.
• Embedded attachments. PDF can carry arbitrary files inside it. Rare, but it does explain the occasional file whose size makes no sense at all.

What a compressor actually does

"Compress PDF" is a bundle of four distinct operations, only some of which are lossless:

1. Downsampling. Reducing image resolution to a target dpi. This is where nearly all the savings come from, and it is irreversible.
2. Re-encoding. Recompressing images as JPEG at a chosen quality. Also lossy — and note that recompressing an image that was already JPEG adds a second generation of artefacts on top of the first.
3. Font subsetting. Discarding unused glyphs from embedded fonts. Lossless for the document as it stands, but it does mean the file can no longer render characters it does not currently use.
4. Structural cleanup. Removing orphaned objects, discarding old revisions, recompressing streams, dropping metadata and thumbnails. Modest savings, no visible cost.

This is why compression results vary so wildly between files. A scan-heavy document might drop by 90% because step 1 has enormous room to work. A text document with no images might drop by 5%, because its streams were already compressed and there is nothing to downsample. If a tool promises a fixed percentage regardless of input, it is describing an aspiration.

Practical fixes, cheapest first

• Re-export from the source. If you still have the original document, exporting again with sensible image settings beats every after-the-fact treatment, because the compressor is working from full-quality originals rather than from something already degraded.
• Resize images before placing them. Scaling a photo to the size it will actually appear before importing it removes the problem at its root.
• Rescan at the right settings. 200–300 dpi, greyscale or bitonal for text. Usually the largest single win available, and the only one that costs nothing in quality.
• Split the document. If the real constraint is a 10 MB email limit, sending three files of 8 MB is often simpler than degrading one file until it fits — and page-splitting is lossless.
• Rebuild the file. Reconstructing a PDF from its pages drops accumulated revisions and orphaned objects. It will not touch image resolution, but on a much-edited document the difference can still be surprising.
• Do not bother zipping it. The bulky parts of a PDF are already compressed streams. Zipping typically saves a couple of percent and adds a step for the recipient.

Frequently asked questions

Why did compressing my PDF make the text blurry?

Because the "text" was a scanned image, and downsampling reduced its resolution. Real text objects are vector instructions and cannot blur at any file size. Blurring after compression is a reliable sign that you are working with a scan.

Is there a lossless way to shrink a PDF?

Yes, but the savings are modest: font subsetting, removing old revisions and orphaned objects, and recompressing streams. Expect single-digit to low double-digit percentages. Large reductions always involve discarding image data.

Why is my PDF bigger than the images I put in it?

Usually because the images were re-encoded losslessly, or converted to a different colour space, or embedded twice. It can also be embedded fonts and structural overhead, which are pure additions to whatever the images weighed.

Does deleting pages reduce the file size proportionally?

Only if the tool rebuilds the file. And the reduction follows the weight of those particular pages, not their number: removing the one page carrying a photograph can shrink the file more than removing twenty pages of text.

What is a reasonable size for a PDF?

For text-only documents, under 100 KB per page. For pages with a few screen-resolution images, 200–500 KB per page. For scans, 100–300 KB per page at 200 dpi greyscale. Anything above a megabyte per page deserves a look.

Further reading

• What is actually inside a PDF file
• Scanned PDF vs text PDF: what OCR does
• Split a PDF into smaller files