PDF Compression: How to Shrink Files Without Wrecking Quality

Almost every PDF that is too large is too large for one of three reasons, and only one of them is really about compression. Understanding which reason applies to your file is the difference between a clean 80% size reduction and a blurry document that is still 12 MB.

Why your PDF is big in the first place

1. Embedded raster images. A phone photo dropped into a report is often 4–8 MB on its own, at a resolution far beyond what any printer or screen needs.
2. Scanned pages. A scanner produces one full-page image per page. A 40-page scan at 600 DPI is a 200 MB file before you have written a word.
3. Embedded fonts and duplicated resources. Fonts embedded in full rather than subset, or the same logo re-embedded on every page instead of referenced once.

Text itself is almost never the problem. A hundred pages of pure text is typically under 1 MB. If your text-only document is 30 MB, something else is riding along inside it.

The three levers compression actually pulls

Image downsampling

This reduces the pixel dimensions of embedded images to a target DPI. For screen reading, 150 DPI is generous. For office printing, 200–300 DPI is plenty. Anything above 300 DPI is invisible to the human eye at normal reading distance and is pure file weight.

Image re-encoding

Photographs compress well as JPEG; line art, screenshots and text-heavy scans do not — they get ugly halos. If your document is mostly screenshots or diagrams, an aggressive JPEG setting is exactly the wrong choice, and you will see it immediately in the edges of text.

Object stream cleanup

Removing unused objects, deduplicating repeated resources, subsetting fonts and compressing the cross-reference table. This is lossless. It typically wins 5–20% and never costs you quality, so there is no reason not to do it.

A practical recipe by goal

• Email attachment under 10 MB: downsample to 150 DPI, medium image quality, lossless cleanup on.
• Web download or portfolio: 120–150 DPI, medium quality. Prioritise fast loading over print fidelity.
• Client deliverable that will be printed: 300 DPI, high quality. Accept a bigger file; a pixellated logo on a printed proposal costs more than bandwidth.
• Archival or legal filing: do not aggressively compress at all. Use lossless cleanup only, and consider PDF/A instead.

The mistake almost everyone makes

Compressing the same file repeatedly. Lossy image compression is generational: each pass throws away information the previous pass already degraded. Two rounds of medium compression look far worse than one round of aggressive compression at the same final size. Always compress from the original, not from a compressed copy.

When compression is not the answer

If your file is a 300-page scan, the real fix is OCR plus a rebuild, not compression — or simply splitting the document so recipients only get the part they need. If your file is large because it contains a full-resolution photographic appendix, extract the appendix into a separate download and link to it. Sending someone a 90 MB file so they can read a two-page cover memo is a workflow problem, not a compression problem.

How to check your work

Open the compressed file at 200% zoom and look at three things: the smallest body text, any thin table rules, and any logo with a gradient. Those three fail first. If they hold up at 200%, they will hold up in print and on screen.

A step-by-step walkthrough for a mixed document

Take a common real-world case: a 45-page proposal with a cover photo, six charts exported as PNG screenshots, and thirty-eight pages of body text. Treating the whole file with one setting is why compression so often disappoints people, because the cover photo and the charts have opposite optimal encodings.

1. Separate the file mentally into three asset classes: the photographic cover, the flat-color charts, and the text pages. Text pages need no image processing at all — verify the body pages are vector text, not a scanned image of text, before doing anything else.
2. Downsample the cover photo to 150 DPI for screen delivery, or 300 DPI if it will be printed, and re-encode as JPEG at a medium-to-high quality factor. A quality factor around 75–85 out of 100 is usually indistinguishable from the source at normal viewing distance.
3. Leave the charts alone or re-encode them as PNG or a lossless mode if the tool supports per-image formats. JPEG artifacts on a flat-color bar chart produce visible speckling around every bar edge, which is disproportionately noticeable because the background is a single flat color.
4. Run the lossless cleanup pass last: subset fonts, deduplicate the repeated logo that appears in the header of every page, and strip the thumbnail cache so it does not carry the pre-compression image data forward.
5. Re-open and check file size versus the 10 MB email ceiling, or whatever your actual constraint is, before deciding whether another pass of downsampling is worth the quality cost.

Benchmark ranges worth knowing

These are typical outcomes, not guarantees, because source material varies enormously. They exist to set expectations before you start, so a result outside this range tells you something about your specific file rather than about the tool.

• Pure scanned document, 600 DPI source, downsampled to 300 DPI: 60–75% size reduction, no visible quality loss for printed text.
• Pure scanned document, 600 DPI source, downsampled to 150 DPI: 85–92% size reduction, visible softening on small footnote text.
• Photo-heavy report, JPEG re-encoded at quality 80 from an uncompressed source: 70–85% size reduction, negligible visible difference.
• Text-only PDF with embedded full font sets, subsetted: 5–15% size reduction, zero visible difference, always worth doing.
• PDF with duplicated embedded logo across 100 pages, deduplicated: reduction scales with duplication count and can exceed 90% on logo-heavy templated documents.

Common mistakes that undo good compression

• Compressing after flattening form fields into rasterized images instead of vector annotations — this converts crisp text into a bitmap that then needs its own DPI settings.
• Applying a single global quality setting to a document containing both scanned pages and native vector pages, over-compressing the vector pages for no size benefit.
• Ignoring color profile mismatches: converting from CMYK to RGB during compression can shift printed colors even though the file gets smaller and looks fine on screen.
• Assuming a smaller file is always a better file — a 2 MB scanned contract that has been compressed into illegibility is worse than a 15 MB scan that a lawyer can actually read.

Troubleshooting specific symptoms

The file barely shrank

Check whether the images were already compressed before you started. A JPEG re-encoded at the same or higher quality than its source will not shrink further; you are compressing a compressed file and the entropy is already gone. Check the DPI of embedded images directly rather than assuming your tool's downsampling triggered — some tools skip images that are already below the target DPI.

Text looks blurry after compression

This almost always means body text pages were rasterized rather than kept as vector text and font glyphs, then downsampled like a photograph. Vector text should never blur under image compression settings because it is not an image. If it blurred, the page was scanned or flattened to an image at some point in the pipeline, and the fix is upstream of compression, not a gentler compression setting.

Colors shifted

Look for an implicit color space conversion in the compression step, particularly RGB-to-CMYK or the reverse. This is a common side effect of some compression libraries normalizing color spaces and is unrelated to the compression ratio itself.

Edge cases worth planning for

• Transparent PNG images with soft shadows: JPEG re-encoding does not support transparency, so a transparent logo re-encoded as JPEG gains an opaque white or black background box that was previously invisible.
• Vector illustrations exported as high-resolution raster images by the originating design tool: these look like photos to a compressor but behave like line art visually, so they need the line-art quality profile even though metadata reports them as JPEG.
• Mixed-orientation scans in one file: a downsampling pass applied uniformly can leave landscape pages under-resolved if the target DPI was calculated against portrait dimensions.
• Password-protected source files: some compression tools cannot read encrypted image streams and will silently skip those pages, leaving the file only partially compressed.

Standards and reference points

The PDF format itself is standardized as ISO 32000-2 (PDF 2.0), which defines the image XObject and stream filter mechanisms — including DCTDecode for JPEG, CCITTFaxDecode for black-and-white fax-style scans, and FlateDecode for lossless zlib compression — that all PDF compression tools operate through. If archival integrity matters more than size, the relevant target is PDF/A, standardized as ISO 19005, which explicitly restricts certain compression and font-embedding behaviors to guarantee long-term reproducibility; PDF/A tooling will refuse or warn on some aggressive compression settings for exactly this reason.

When not to use aggressive compression

Do not aggressively compress documents that will go through a further generation of processing, such as OCR, print prepress, or legal exhibit stamping — each of those steps works better on a higher-fidelity source, and you can always compress the final output once. Do not compress medical imaging or engineering drawings where diagnostic or measurement accuracy depends on pixel-level fidelity; use lossless-only settings there regardless of file size. And do not compress a file you are about to sign digitally in the same batch as other edits, since re-saving after signing invalidates the signature anyway — get the file to its final compressed form first, then sign.