Split a PDF into the pages you need
Pull out a page range, drop the pages you do not want, or cut a long document into blocks. The pages come across untouched, and the file never leaves your device.
Pick a PDF and its page count, page size and version are read out below.
Document
One new PDF containing exactly these pages, in the order you write them. 3,1,2 reorders.
Five different things people mean by “split”
The word covers several jobs that have nothing to do with each other, which is why so many tools do only one of them and feel broken. All five are here:
Keep only these pages pulls a selection into a single new document — the usual case when someone asks for “just the signature page and the schedule”. The order you write the pages in is the order you get, so this doubles as a way to reorder a document. Remove these pages is the inverse: everything survives except what you list, which is what you want for stripping a cover sheet or a blank scan.
One file per page explodes the document completely, numbered and zero-padded so they sort correctly in a file manager. Every N pages cuts it into equal blocks — useful when a portal accepts uploads of at most ten pages at a time. Split at page numbers is for documents that are really several documents stapled together: give it the page each new section starts on and you get one file per section.
Writing the page range
A range is a comma-separated list. 1-5 is a span, 2,7,9 is a list, and the two mix freely: 1-3, 8, 12-15. An open end works in both directions — 10- means page ten to the end, -4 means the start through page four. A reversed span such as 9-6 is read as the span between those pages rather than treated as an error.
Numbers that do not exist are reported rather than ignored. Asking for 1-40 in a twenty-page document tells you so and takes the twenty that exist, instead of silently handing back a shorter file than you asked for; that silent behaviour is how people end up emailing an incomplete document.
These are positions in the file, counting the first sheet as 1. If the document has a cover and a contents page before the printed “page 1”, the numbers will not line up with what is printed, and the summary above the controls tells you how many pages there really are.
Why the parts do not add up to the whole
This surprises people, so it is worth stating before you wonder whether something went wrong. Split a 4 MB report into one file per page and the total will usually be several times more than 4 MB. Measuring 139 assorted real-world PDFs while building this page, the one-page files added up to a median of about 3.3 times the original, and only three of the 139 came out smaller than they started.
The reason is that a PDF shares its resources across pages. One embedded font subset, one logo image and one colour profile are stored once and referenced by every page that uses them. As soon as each page becomes its own document, each document needs its own copy of everything its page refers to. A page with three lines of text on it can easily carry a 200 KB font subset, because that is the subset the whole report shared. Files produced by browser print-to-PDF and by JavaScript PDF libraries are the worst offenders, because they tend to embed whole fonts rather than subsets.
The practical consequences: splitting to get under an upload limit works well for blocks of pages and badly for single pages, and if you need genuinely small single pages, the font is usually what to attack, not the page count. Splitting into a handful of blocks stays close to the original total, because each block amortises its fonts over many pages.
Nothing is re-compressed
Each output page carries the same content stream bytes as the page it came from, with the same compression it already had. Scans are not re-encoded, so they do not get blurrier; text stays real text and stays searchable; vector drawings keep their full precision. Page size and rotation are properties of the page and travel with it, so splitting a document with a landscape appendix gives you an appendix that is still landscape.
Form fields come across with their field tree, so a page cut out of a fillable form is still fillable. Links to web addresses keep working. Links that jump to another page work when that page is in the same output file, and are dropped when it is not — a link pointing at a page that is no longer there would be worse than no link.
Bookmarks are not carried over. Splitting by bookmark is a genuinely useful feature and this page does not do it: reading the outline tree is easy, but deciding which bookmark levels define a section is guesswork, and guessing wrong quietly produces the wrong files.
Protected and damaged files
A file with an owner password — one that opens without a prompt but is flagged no-printing or no-copying, which is what banks and payroll systems almost always send — is read normally, and the pieces come out without the restriction. A file that asks you for a password when you open it is refused, and no attempt is made to get past it. Unlock it in a reader first and split the unlocked copy.
If the file's cross-reference table is damaged, which happens to interrupted downloads, the whole file is scanned for object headers and the index rebuilt from scratch. That often recovers a document a reader refuses to open at all. The summary says when this has happened, so you know to check the output rather than trust it.
Common questions
How do I get just one page out of a PDF?
Choose Keep only these pages and type that one page number. You get a single-page PDF with the page exactly as it was — same size, same rotation, same image quality, still searchable. If you want several pages as separate one-page files instead of one file containing them all, use One file per page.
Can I change the page order while splitting?
Yes. In Keep only these pages, the order you write the numbers is the order you get. Typing 3,1,2 produces a three-page document in that sequence, and 5-1 is read as the span from 1 to 5 rather than a reversal, so list the pages individually when you want them backwards.
Why are my split files bigger than I expected?
Because fonts, logos and colour profiles that the original shared between pages have to be copied into every file that needs them. A single page pulled out of a report carries its own copy of whatever font subset that page uses, which is often far larger than the page's own text. Across 139 assorted real documents, splitting to one file per page produced a total about 3.3 times the original on average. Splitting into blocks of several pages stays much closer to the original size. This is inherent to the format, not something the tool is doing wrong.
Does splitting reduce quality?
No. Page content is copied byte for byte with its original compression, so scans are not re-encoded and vector artwork keeps full precision. The only thing that changes is the structure around the pages: a new catalogue, page tree and cross-reference table.
What about a password-protected PDF?
If it opens without asking you for a password — the common case, where only an owner password is set to block printing or copying — it is read normally and the output has no restriction on it. If it prompts for a password to open, it is refused. Unlock it in your reader, save a copy, and split that.
Can it split by bookmark or by chapter?
No, and that is deliberate. Reading the outline tree is straightforward, but working out which bookmark level marks a chapter boundary is a guess, and a wrong guess produces plausible-looking files cut in the wrong places. Split at page numbers does the same job with you making that decision.
Is there a size limit?
No fixed one. The file is read into your browser's memory, so the ceiling is whatever your machine will give the tab — comfortably into the hundreds of megabytes on a desktop, less on an older phone. Nothing is uploaded, so there is no upload limit, no queue and no daily cap.
Do the output files keep the original filename?
They are named after it. One file per page gives you report-page-01.pdf and so on, with the numbers zero-padded so they sort correctly; block modes are named with the page range they contain. When there is more than one output you can take them individually or as a single ZIP.