How to split a PDF on any device

Four methods that work, and what each one did to 34 real documents.

The short answer

On a Mac, drag the pages you want out of Preview's thumbnail sidebar. On Windows, print the page range to Microsoft Print to PDF. In a browser, use a tool that cuts the file in the page rather than on a server.

Then check two things in what came out. Search it for a word you know is on those pages, because a tool that rebuilds the document can drop the invisible text layer while the pages still look perfect. And look at the size, because splitting always costs something and one method below cost a lot.

On a Mac, Preview already does it

No download, no account, and it works with the network off.

1. Open the PDF in Preview

   Double-click it, or right-click and choose Open With, then Preview.

2. Show the thumbnails

   View, then Thumbnails. The pages appear down the left edge.

3. Select the pages you want

   Click one, then hold ⌘ to add more or ⇧ to take a run.

4. Drag them to the desktop

   Dragging the selection out of the sidebar writes a new PDF holding those pages. The original is untouched.

This is the fastest route on a Mac and it is the one worth checking afterwards. The framework underneath it rewrites the document rather than copying it, which is where the size and text results further down come from.

On Windows, printing is the only built-in route

Windows 11 has no Save selected pages command. What it has is a virtual printer, and printing a page range to it produces a file holding that range.

1. Open the PDF in Edge

   Or in the Reader app, or in any viewer with a print dialogue.

2. Press Ctrl and P

   Choose Microsoft Print to PDF as the printer.

3. Type the pages

   Under Pages, enter something like 1-3 or 2,5,9.

4. Print, and name the file

   Windows asks where to save it. That saved file is your split.

One pass gives you one file. Cutting a 40-page report into four parts means four trips through the print dialogue, which is the reason most people reach for something else.

In a browser, without the upload

A browser tool is the only route that is the same on every operating system, and some of them do the work in the page rather than on a server.

That distinction is worth checking rather than trusting. Of the tools ranking on the first page for "split pdf" on 13 September 2026, iLovePDF, Smallpdf and Adobe all sent the document to a server and served the result back from a download token. pipefile and ihatepdf.cv did the work locally. A browser tool that never uploads also keeps working with the network off once the page has loaded, which is a quick way to test the claim.

This site's own splitter is Split PDF. Its results are the floi rows in the table below.

On the command line, when there are fifty of them

Two page ranges is a mouse job. Two hundred files is not, and this is where a shell loop pays for itself.

Poppler's pdfseparate is measured below. The qpdf line is written from its documentation and is not in the table, because qpdf was not installed on the test machine and this site does not publish numbers it did not run.

What each method did to 34 real PDFs

Every document was split in half, three ways, on 13 September 2026. The corpus is 34 PDFs from the author's own Downloads folder, 108 pages and 58,559,349 bytes in total, from a 24 KB bank statement to a 28.7 MB scan. They are not committed to the repository, because they are personal files. Any folder of PDFs reproduces the shape.

Machine: macOS 26.6.2 on Apple silicon, Node 24.18.0, poppler 26.08.0, PDFKit as shipped with the OS. Text was extracted from the source and from the two halves with pdftotext and compared character for character after whitespace normalisation.

Every method comes out above 100%, and that is the format rather than the tool. Splitting a document in two puts the fonts and images both halves use into both halves. The interesting number is how far above 100% a method lands: copying the page objects costs 1.4%, and rebuilding the document cost 76.5%.

The PDFKit result

PDFKit is the framework macOS Preview and the Finder's Quick Actions are built on, so it is the engine behind the Mac route at the top of this page. It changed the extractable text of 6 of the 34 files, and it grew several of them severely.

On a two-page agreement of 45,410 bytes, the two halves came to 1,204% of the original. The pages render identically: what changed is how the document is written, not what it draws. Five of the six text failures were machine-generated invoices and statements, which is exactly the kind of document people split.

The one floi difference

Freelancer Annotation Onboarding Guide_FINAL.pdf extracts the same characters in a different order after a floi split. It is a tagged PDF, and the structure tree is not carried into the parts, so pdftotext falls back to geometric order. Poppler carried it and floi did not, which is the row where floi comes off worse and it is stated on the tool page as a limit.

Why your one extracted page is still enormous

This is the complaint behind most of the searches, and the answer is the same shared resources. Page 1 was pulled out of each of the same 34 documents and weighed against the whole file.

Median share across all 34: 41.6%. The pattern is the content, not the page count. A text document's weight is its embedded fonts, and every page carries them, so one page of two is nearly the whole file. A document of photographs keeps its weight in the images, and each page only takes its own.

No splitter can fix this by cutting more carefully, because the page genuinely needs those fonts to draw. If the extracted page has to be small, compress it afterwards or export it as a picture with PDF to Images.

Which one to use

Reproducing this

The scripts are a dozen lines each and are described in docs/research/split-pdf.md in this site's repository, alongside the method, the machine and the date. They are not committed, because they read private files.