Extract PDF Pages handles PDFs up to 50 MiB and 500 pages, copying only the pages you specify into a new downloadable file that never leaves your browser. Files beyond either boundary are rejected rather than silently trimmed, so the maximum file size and page count for extraction are hard ceilings rather than soft targets. Reading the source, parsing your selection, copying the relevant page objects, saving the result, and preparing the download all happen locally — your file, your selection, and your output remain on the current device. If your PDF is heavier than 50 MiB or longer than 500 pages, you'll need a different approach before this tool can help. Inside those limits, output order is controlled by token order, and duplicate pages are kept on their first appearance while later occurrences are reported.
What "Large" Usually Means for a PDF
PDFs get "large" in three different ways, and each one affects extraction differently. The most common cause is many pages: a 200-page scanned book or a 400-page reference manual can weigh only a few megabytes per page, yet feel unwieldy when you need only a handful of sections. The second cause is heavy embedded resources on a smaller page count — high-resolution photographs, vector drawings, custom font subsets, or scanned layers — which can push a 30-page document past 100 MiB. The third cause is repetition: a directory of nearly identical forms, a packet of attachments, or a portfolio-like document where most of the bytes live in shared resources rather than unique pages. Understanding which kind of "large" you have matters, because the tool's limits are stated in both file size and page count, and a single 20 MiB page can blow through the byte budget while leaving you well within the page budget.
How Extract PDF Pages Handles Big Files
Extract PDF Pages applies two hard ceilings before any work begins: the input file must be 50 MiB or smaller (52,428,800 bytes exactly), and the decoded document must contain between 1 and 500 pages. A file at either boundary is accepted; a file that exceeds either is rejected outright, not shortened. The browser checks the PDF MIME type or extension and the byte limit before pdf-lib's PDFDocument API loads the source without bypassing encryption, which means password-protected files cannot be opened here even if you know the password. Because every step runs locally — reading the bytes, parsing your page expression, copying selected page objects into a new document, saving the result, and exposing it through a temporary object URL — there is no upload, which makes the tool practical for large internal documents you would not want to send to a remote server. the related extraction guide.
| Boundary | Value | Behavior at the limit |
|---|---|---|
| Maximum input file size | 50 MiB (52,428,800 bytes) | Accepted; larger files rejected |
| Maximum page count | 500 pages | Accepted; longer documents rejected |
| Selection field length | 4,000 characters | Over-limit input is rejected, not truncated |
| Encryption | Not bypassed | Password-protected files cannot be opened |
| Processing location | Browser only | No upload of source, selection, or output |
Extract Pages From a Large PDF (Step by Step)
Follow these steps to pull specific pages out of a PDF that sits within the 50 MiB and 500-page envelope.
- Open Extract PDF Pages in your current browser and click the file picker to choose the local PDF. Confirm the file is unencrypted, 50 MiB or smaller, and contains 500 or fewer pages; otherwise the browser will refuse to load it.
- Type your selection into the page field using one-based page numbers and inclusive ascending ranges, separated by commas or whitespace. For a large manual where chapter seven starts on page 47, an entry like 47-58, 5, 1-3 means: take pages 47 through 58 in original order, then page 5, then pages 1 through 3 — total 16 pages.
- Keep your expression under 4,000 characters; anything longer is rejected in full rather than silently truncated. If your selection is huge, split it into multiple runs and concatenate the resulting PDFs afterward.
- Run the extraction. The tool validates each token, discards descending ranges like 5-3, rejects decimals, negatives, and malformed ranges, and flags any page that falls outside the loaded document with an explicit error.
- Read the duplicate notice if one appears. The parser keeps each page's first occurrence, so 3, 1-3 produces page 3 first, then page 1, then page 2 — the second time page 3 appears in the input, it is ignored and the duplicate is reported in the result.
- Download the new PDF from the temporary link. Loading a different file or re-running extraction replaces the previous output; closing the page releases the temporary URLs entirely.
- Open the downloaded file in your target viewer, walk through the page order, and confirm that fonts, images, and any links you rely on render the way you expect before sharing or printing.
Output Size, Memory, and Resource-Heavy Pages
A common surprise with large PDFs is that the output does not scale in direct proportion to page count. Selected page objects carry their own resources — fonts, images, color profiles, form XObjects — and those resources are copied into the new document along with the page. Extracting a single page from a catalog that embeds a 30 MB image can therefore produce an output that is only marginally smaller than the source, even though the page count drops by 99 percent. The opposite is also possible: removing repeated resources from a 400-page deck can shrink the file dramatically because shared fonts and images stop being referenced. Browser memory availability varies by device and by tab workload, so a 50 MiB file that loads fine on a desktop may struggle on a low-RAM laptop or an older phone. If extraction fails partway through, the temporary result link is replaced on the next run rather than left dangling, and the source link to the original local file is the only persistent reference to your input. Closing the page releases all temporary URLs.
Duplicates, Ranges, and Parser Rules
Because the parser is the same one used in production and tests, the behavior is deterministic rather than best-effort. Tokens are processed in the order you type them, and that order becomes output order, so 5, 2-3 produces a PDF whose first page is the original page 5, whose second and third pages are the original pages 2 and 3. The duplicate rule is "first occurrence wins": if you list 3, 1-3, 3, the result is page 3, page 1, page 2 — the third occurrence of page 3 is reported as a duplicate and ignored. Descending ranges like 5-3 are rejected with an explicit error rather than reversed to 3-5, and the same is true for decimal values, negative numbers, malformed ranges, and any page number that exceeds the loaded document. This strictness is intentional: it prevents the silent reordering that would otherwise turn a typo into a wrong deliverable. If you actually need duplicate pages in the output, use a workflow designed for that — the tool is built to deduplicate, not to clone. The production extractor loads the original bytes, validates the zero-based unique indices, creates a new PDFDocument, copies pages in the parsed order, saves it, and exposes the new bytes through a temporary local object URL.
What Survives Extraction and What Doesn't
Extracting pages is a structural change to a PDF, not a lossless copy. The tool copies each selected page's object — including its 90-degree rotation, crop box, media box, page graphics, and ordinary page resources — without rasterizing the page into a screenshot, so visual fidelity is preserved. Several features, however, depend on document-wide relationships that do not travel cleanly with a subset of pages. Digital signatures will not remain valid in a different PDF, so any signed page extracted here is no longer signed. AcroForm fields, annotations, scripts, embedded files, layers, page labels, bookmarks, document outlines, cross-page destinations, and links can reference objects outside the selected pages and may behave unpredictably in the output. Dynamic XFA forms are not a supported preservation target at all, per the underlying pdf-lib library's encryption and XFA handling. The tool does not promise that every interactive, navigational, signed, or archival property survives — it makes a new, valid PDF that contains the pages you asked for, and you are responsible for confirming the result in your target viewer. For documents where conformance, signatures, or accessibility structure matter, retain the original PDF and verify with a specialist editor.
Checking the Result Before You Rely On It
Because the tool reports duplicates and rejects bad input but does not validate interactive features, the final check is on you. Open the new PDF in the viewer your recipient will use, walk the page order against the original page numbers, and test any link, form field, or annotation that matters to your workflow. Compare dimensions and orientation against the source using the local source link the tool exposes for that purpose. If the output is bigger than you expected, that is a resource-copying effect rather than a bug: the new file contains only the pages you selected, but those pages still carry the fonts and images they need. A real integration test in the tool builds source pages with different sizes and identifiers, extracts them through the production function, saves the result, reopens it, and verifies page count, output order, dimensions, and identity — catching implementations that merely return the right number of pages while copying the wrong pages or sorting the user's order. If a feature you depend on does not survive, the original PDF remains on your device; nothing was uploaded, so there is no remote copy to clean up.
Related reading: Extract PDF Links Free, Unlimited, in Your Browser.