Separate positions or ranges
Use commas for separate pages and a hyphen for an inclusive range. For a five-page source, 1, 3, 5 keeps three pages; 2-4 keeps the middle three.
Extract pages from PDF
Keep the pages you need in one new PDF, in their original order.
No PDF selected
Choose a PDF to select its pages.
One PDF · 1–200 pages · up to 50 MiB · one new PDF download
Choose one PDF to start.
Review and select original pages
Document labels are shown when present, but the range field always uses original physical positions.
Your selected pages
In your browserNo uploadsNo account
How to extract pages from a PDF
- Choose one PDF
Select a local document and wait for its file check. The original file is not overwritten.
- Keep the pages you need
Enter positions such as 1, 3-5, or open the original-page preview and toggle a page. Review the selection summary.
- Check the new document
Extract, preview the result and download the new PDF. Open it to check the pages before sharing.
Choose a subset, keep its order
One copy of each selected page
Input such as 5, 2, 2-3 produces original positions 2, 3, 5. Pages are sorted into source order and overlaps appear once. This tool does not reorder or duplicate pages.
Positions, not printed labels
A cover labelled i is still physical position 1. The preview selector can display document labels, but the range field uses positions. A single-page source is accepted with selection 1.
Support and limits
Copied pages, not screenshots
The tool copies selected page objects into a new PDF. Text pages remain text; scanned pages stay images without OCR. Page size, crop and rotation need review in your result. HTTP, HTTPS and email link targets are allowed.
Whole-file checks and resource limits
One PDF up to 50 MiB, 1–200 original pages and 20,000 PDF objects; output up to 100 MiB. Checking and extraction each use a 60-second timer, including initial preview. Background browser throttling can delay timer delivery. Cancel stops the current operation and removes unfinished output. Memory may impose a lower practical limit.
Document features that stop extraction
Encrypted or damaged documents, forms, signatures, certification restrictions, comments, non-link annotations, internal navigation, layers, portfolios, attachments and document actions are rejected. The check applies to the entire input, including pages you do not select. Bookmarks, page labels, document metadata and accessibility tags are not rebuilt.
A focused copy, not redaction
Other page positions are omitted from the new document, but hidden text or shared resources on selected pages are not inspected or cleaned. This is not a redaction tool, file-size target, signature-preservation workflow or PDF/A/PDF/UA compliance service. Keep your original.
Test files and results
Use these original English fixtures to compare the visible source pages with the extracted PDF. Each row describes its exact selection, not every possible PDF.
| Source and selection | Actual PDF download | Observed result |
|---|---|---|
| Five-page mixed PDF · 1, 3, 5 | Three non-adjacent pages | Three pages in original order: the report, rotated/cropped appendix and memo. Extracted text, page content, boxes, rotation, image data and external links matched the selected source pages; rendered pixels matched at 90 dpi. |
| Same source · 2-4 | Middle three pages | Three pages in original order: the second report page, rotated/cropped appendix and image-only receipt. The receipt stayed image-only. The same page-content and 90 dpi pixel checks matched the source. |
| Labelled five-page PDF · 5, 2, 2-3 | Normalized subset | Original physical positions 2, 3 and 5 appeared once, in that order. Document labels ii, 1 and 3 did not replace physical positions. The same page checks matched; original document labels were not rebuilt. |
Download the page-by-page check records.
What these checks cover
On October 28, 2026, we downloaded these three PDFs and viewed all nine result pages. We compared each selected source page with its output text, content, page boxes, rotation, images, external links and rendered pixels at 90 dpi. A separate single-page input and a 390-pixel browser-width selection/conversion/download also completed. This was not a physical-phone test.
Device and document scope
Desktop and 390-pixel browser-width checks are distinct from a physical-phone test. Results on these fixed files do not establish every device's memory, speed or document compatibility. Refused inputs produce no completed download.
Local processing and privacy
Your document stays local
The input file, its name, selection and new PDF are handled in browser memory, without uploading to a conversion server. Clear the file or close the page to release the working state. Downloads remain on your device.
Separate website requests
Code and PDF-rendering assets load from this site. Website delivery and analytics are separate requests described in the privacy policy; file names, document contents, selections and results are not sent as analytics events.
Frequently asked questions
Why are my extracted pages not in the order I typed?
The result follows the original PDF order. Repeated pages and overlapping ranges appear once; the selection summary shows the exact set. This is a subset extractor, not a page-reordering tool.
Why did an unselected page cause the file to be refused?
The input must pass a whole-document feature check before extraction. Unsupported signatures, actions, annotations or other document structures can exist outside the selected pages.
Does extracting a page recognize text in a scan?
No. The selected page is copied as it is. An image-only page stays image-only; its visible content can be kept without OCR.
Do I enter the page number printed on the document?
Enter the physical position counted from the beginning of the PDF. Printed numbers and document labels can differ; check the original-page preview before extracting.
Why did the download disappear when I changed my selection?
The old file reflected the previous selection. Extract again to create a result matching the current page set.
Page updated: