Split a PDF
Cut one PDF into several files by page ranges, every N pages, or one file per page.
How it works
Splitting answers a question the file itself cannot: where one document ends and the next begins. A scanner hands back forty pages that are really twelve invoices; a board pack arrives as a single PDF when four people each need one section of it.
Three ways to place cuts
There are three ways to say where the cuts go. Chosen ranges gives one file per comma-separated range, so 1-3,7,10- produces three documents and leaves out pages 4 to 6 and 8 to 9 entirely. They are not deleted from anything, since the file you selected is never modified, but they will not appear in any part either.
Every N pages walks the whole document in equal chunks and puts the remainder in the last file. One file per page does what it says. Page numbers are 1-based throughout: page 1 is the first page, the number your reader shows in its own page box, not an index counted from zero.
Duplication and the rebuild cost
The download is often bigger than the file you started with, and it is worth knowing before you see it rather than after. Two costs stack. The first is duplication: a PDF stores fonts, colour profiles and images once and points at them from every page that uses them, so each part needs its own copy of everything its pages reference.
The second is the rebuild itself, and it runs in both directions. Copying every page of a document into a fresh one while changing nothing moved eight real test files anywhere between 40% smaller and 109% larger. It grows where the source packed its cross-reference tables into compressed object streams, which a rebuild writes out plainly again. That produced the two worst cases, 109% on a 127-page print PDF and 63% on a 138-page one.
It shrinks where the source was still carrying objects nothing references, which a rebuild drops: a two-page government form came out 40% smaller and a slide deck 15% smaller. Split those two print PDFs into five parts each and the parts total 150% and 117% more than the original.
What a cut cannot carry
Two things cannot survive a split intact, and no splitter can change that. The bookmark tree is stored once at the document level, above the pages, and describes a structure the parts no longer have, so it is not carried across and each part opens with an empty outline. Internal links behave the same way: a cross-reference on page 2 pointing at page 30 has no meaning once page 30 lives in a different file, and such links go inert rather than jumping somewhere wrong.
Interactive form fields go the same way for the same reason: the form's definition sits in the document catalogue, not on the page, so a filled field keeps its appearance and stops being fillable, arriving as a flat box with the value drawn into it.
What travels with each page
Everything that genuinely lives on a page does come across: text still selectable as text, vector artwork still vector, images at their original resolution and compression, and annotations. Page geometry comes across too. A rotated page arrives rotated, and a print-ready page keeps the smaller visible box that hides its trim and bleed, so the parts open looking the way the original did. Document metadata such as title and author is not propagated; the parts are new documents.
Why you get a ZIP
A split into eight is eight downloads, and a browser hands you one file at a time, so anything producing more than a single document comes back as a ZIP. Each part inside is named for the pages it holds and numbered with a leading zero where needed, so the archive lists in document order in every extractor rather than putting part 10 before part 2.
The archive also takes back a good part of the growth described above, because what the rebuild adds is redundancy and redundancy is what compressors eat: the two print PDFs whose parts totalled 150% and 117% more than the original ship as archives 47% and 54% larger, and several test files came out at or below their original size. Unzipped, the parts are still the larger figure; run them through compress PDF if that matters.
When your settings produce exactly one document you get that PDF directly, with no archive to unpack. Parsing, page copying and the archive are all built in a Web Worker on your own machine by a WebAssembly PDF library, which is why a 200 MB file has no queue and no upload bar.
Questions
Is my PDF uploaded?
No. The file is read into memory by this page, split by a PDF library running in a Web Worker on your machine, and handed back as a download. Open developer tools, switch to the Network tab and split a document: you will see the code chunks the page loads for itself and no request carrying your file.
Why do the parts add up to more than the original?
Two reasons. Resources shared between pages have to be duplicated, so one font subset serving forty pages becomes forty copies if you split into forty files. Separately, rebuilding a document re-serialises its structure: where the source packed its cross-reference tables into compressed object streams a rebuild writes them out plainly, which made a 127-page print PDF 109% bigger before any splitting happened. That cuts both ways: three of eight real test files came out smaller on the rebuild alone, because a rebuild also drops objects nothing references any more. The ZIP wins much of the growth back, since it is redundancy rather than lost compression, and compress PDF wins back more.
What happens to pages I do not put in any range?
They are left out of every part. The file you selected is untouched (nothing here writes to it), but a page you want in the output has to be inside a range. If you want the whole document covered with no gaps, use the every N pages mode instead.
Can I split an encrypted or password-protected PDF?
Usually yes, and the distinction matters. Most encrypted PDFs (tax forms, bank statements, e-books) are locked against printing or editing but open for reading without anyone being asked for anything, and those split normally. Of 516 real documents tested here, all 8 carrying encryption were this kind. The exception is a file that demands a password before it will open at all: that one is refused, because the pages genuinely cannot be read, and this tool will never prompt you for a password. Asking for one on a site whose whole claim is that it receives nothing would be absurd. Open it in the reader that has the password, save an unprotected copy, and split that.