Extract PDF pages

How it works

This is the tool for a sentence like "I only need pages 4 to 6 of this". You mark the pages worth keeping and everything unmarked is absent from the result: one file, containing your selection and nothing else. It is the same arithmetic as deleting pages, phrased from the other end, and it is a different job from splitting: splitting cuts a document into a set of files, extraction lifts a subset out into a single one. If you want each chosen range as its own document, use the splitter instead.

Pages come out in document order whatever order you clicked them in, and a page named twice appears once; if you need a page repeated, or the pages in some order other than the one they already have, arrange them after extracting.

Pages are copied verbatim

The pages are copied, not redrawn. A PDF page is an object holding a content stream, a set of resources it points at and a list of annotations, and extraction copies those objects into a new document verbatim. Nothing is rasterised and nothing is re-encoded.

A paragraph therefore arrives as characters, still searchable and still selectable; line art stays resolution-independent and prints at whatever your printer can manage; a photograph keeps the exact pixels and the exact compression it had in the source, so the copy is not a generation worse than what you started with. That is the practical gap between extracting and the widespread habit of printing a page range to a new PDF, which turns living text into a flat image of itself and takes every link on the page down with it.

Why the saving is small

It will be smaller than the original, but rarely in proportion to the pages you kept. Keeping the first 10% of the pages of eight real documents gave anywhere from 2% of the original size to 62% of it, clustering around a third. Two things stop the saving being proportional. The pages you keep drag their references along (fonts, colour profiles and images that were shared with the pages you dropped), while material only the discarded pages used goes away.

The second reason is that the new file is written from scratch rather than trimmed, and that rewrite can go either way: a document whose index was stored in packed form loses the packing and gains bytes, while one dragging objects nothing points at any more sheds them and loses bytes.

Across the same eight files that step alone, with every page still present, landed between 40% below the original and 109% above it. So a small extract from a large source is not a dependable way to get a small file, which is why compress PDF sits next to this one rather than as a generic suggestion.

What the extract leaves behind

Three things do not come across at all. The bookmark tree is a document-level structure describing a table of contents the new file no longer matches, so it is not carried over.

Internal links whose destination was not part of your selection cannot survive: a jump to page 30 has nowhere to land once page 30 is gone, so the link goes inert rather than leading somewhere wrong, though links to external addresses are unaffected. And an interactive form field keeps its appearance, so a filled form still reads correctly, but a form's registry belongs to the document as a whole rather than to any single page, so the extracted copy is a flat picture of the filled form and can no longer be typed into.

Page numbers mean file position

Pages are counted the way your reader counts them: page 1 is the first page in the file. That is worth stating because it is often not the number printed on the paper. A report with roman-numeral front matter can have an "iv" that is the fourth page of the file and a "1" that is the eleventh, and this tool always means position in the file.

You can point at the pages on the thumbnails or type the range and the two stay in step, because on a four-hundred-page document typing 200-215 beats scrolling to find it. All of it runs in a Web Worker on your own hardware through a WebAssembly PDF library, so a confidential document is read, copied and written back without a byte of it going anywhere.