A Project Gutenberg book is free of US copyright, and the US is the only country the project makes a promise about. On American soil you may copy the text, change it, sell it or perform it, provided you either strip every mention of Project Gutenberg from the copy or follow the project's license for the name. Anywhere else you have to check your own country's law first. Links and downloads have their own rules: link to the book's landing page, and fetch many books from a mirror or from the catalog files, never by pointing a program at the main website. This guide walks through each of those four points with the wording of the project's own pages, so that a reader, a website owner or a script author knows where the limits are.

What "free" means on this site

Project Gutenberg is run from the United States and applies United States copyright law. The permissions page says that a work that is free there may still be protected abroad, and a work free abroad may be protected there, and it advises anyone outside the country to get professional guidance before redistributing (Project Gutenberg permissions page). The terms of use repeat it for readers: people outside the United States are told to check their national law before accessing the books, and the project offers no warranty about rights anywhere else (Terms of Use).

The wording is more careful than most summaries of it. According to the license explanation, the project dropped the phrase "public domain" from its own license, because readers interpret it in many different ways and then disagree about who may do what (license explained). The page speaks of books that are not restricted under US copyright law instead. This guide follows that habit when it can, and uses the common phrase only as shorthand.

Two more details matter. First, the catalog can be wrong about a book's status; the license page says to look at the notice inside the ebook itself (license explained). Second, a copyright flag is not a quality flag and not an edition flag: the project is the publisher of its own ebooks, which are made with changes such as modernised spelling, so a text is rarely identical to one printed edition (permissions page).

Two parts in one file

The license page describes every ebook of this kind as two things welded together: the book text, which is free of US copyright, and the Project Gutenberg license and trademark, which are not. Remove the license and every reference to the project, and what remains is a text without restriction in the United States (license explained). The trademark side applies only if you keep the name attached. The project lists its conditions for that case:

  • You distribute verbatim copies. Reformatting into another file format is allowed, edits to the content are not.
  • If you charge money for copies, you owe royalties. The full text names 20% of gross profits, calculated the way you already calculate taxes, with payment due within 60 days after the relevant tax return (full license, paragraph 1.E.8).
  • You refund customers who disagree with the license, or who received a defective copy.

A mention of the project in an acknowledgement, a reference list or a colophon is not a use of the trademark, and the project asks only that such mentions carry a link (license explained). Quotations from a book need no permission, even in a commercial product, and nobody has to cite the project as the source, although a citation is welcome (permissions page).

When the header stays and when it goes

The decision depends on what you do with the file. This table is our reading of the pages above; the license text is the authority.

What you do What the license asks Where it says so
Give away an unchanged copy, free Keep the full license with the file and show the standard sentence prominently paragraphs 1.E.1 to 1.E.6 of the full license
Sell copies and keep the name Pay the royalty, offer refunds, keep the license paragraph 1.E.8 of the full license
Change the text, build a derivative, or sell it without royalties Remove the license and every reference to the project, in the United States paragraph 1.C of the full license
Mention the project as a source Nothing, a link is appreciated license explained

"Every reference" is the demanding part. It means the header, the footer, the license block and also the project's name wherever it appears in the text body, such as a transcriber's note. A copy from which only the long license block was deleted still advertises the project and is therefore still bound by the terms for the name. A check that works: after cleaning, search the whole file for the word "Gutenberg" in any capitalisation. Any hit means the copy is not clean. Run it on a file you know still has the header, to see the search can find something, before you trust the empty result on the cleaned one.

Linking to a book

The project asks for links to the landing page of a book, for example the /ebooks/ page with its number, rather than to a particular HTML, text or EPUB file. Deep linking to files is discouraged by technical means, using the referrer, because the back-end layout has been renamed several times and the landing page is meant to be permanent (Terms of Use). If you must point to a file or a passage, the terms say to host your own copy.

Two limits follow for anyone who builds a site. Large-scale deep-linking to files on the project's servers is not allowed, and the project reserves the right to act against it technically and legally; mock front-ends that earn advertising while the project pays for the bandwidth are named as unwelcome (Terms of Use). A page that wants to send readers to a book should link the landing page and credit the project, and let the reader's own click start the download. The pages we read do not say how the numbered download addresses under /ebooks/ are treated when they are placed on another site, so we did not test that and do not claim anything about it.

Downloading without being blocked

The main website is for human users. The terms and the robot policy both say that any perceived automated use leads to a temporary or permanent block of the IP address, and that blocks are applied automatically from traffic volume and expire after a few days (robot access policy). Anyone behind a shared proxy or VPN can be caught by somebody else's traffic.

The footnote in the terms defines "many books" as more than about 100 per day, and sends anyone beyond that to a mirror (Terms of Use). The options the project documents:

  1. For a local copy of everything, set up a private mirror; the mirroring guide recommends rsync, because wget and cURL must inspect all files to find the recently changed ones (mirroring how-to).
  2. For a selection, the robot policy offers a harvest address that is meant to be used with a wait of 2 seconds between requests and a filter for file type and language (robot access policy).
  3. For a list of books and their metadata, use the machine-readable catalog, not the website. The offline catalogs page offers an XML/RDF file refreshed daily and a CSV file refreshed weekly, and it asks developers to use these instead of crawling the site (the offline catalogs and feeds page, listed among the sources). The plain-text GUTINDEX lists are for reading by eye and are not recommended as input for automation.
  4. For an app that reads the OPDS feed, send a proper user agent with a contact address and make no more requests than a person with a browser would; high-volume apps should host their own feed from the catalog (Terms of Use).

The robot policy adds a courtesy that is easy to miss: generated formats are rebuilt monthly and hundreds of books are corrected each month, so a saved copy should be refreshed from time to time (robot access policy). For a similar case where a block on automated clients was misread as a result about the site, see our note on crawler tests that misread a block.

What a download count means

A list of books ordered by downloads looks like a measure of readers. The project's own top page defines the counting more narrowly: several downloads from one internet address on one day count once, and addresses that fetch more than 100 books a day are treated as robots and not counted (top downloads page). The same page publishes the counts for yesterday, the last 7 days and the last 30 days, which already shows that one book can sit in different places depending on the window.

Gutendex, an interface to the catalog, orders books by popularity from the number of downloads and returns a field named download_count (Gutendex). Its page does not state which time window that number covers, so a count copied from it is a snapshot of one source at one moment. Treat a count as a rough rank, never as praise, and keep the date of the read next to it.

When this does not apply

Policy pages change, and what we wrote is a plain reading of them as of 30 October 2026, never legal advice. Nothing here settles whether a particular book is free in your country; for that, ask a lawyer or the copyright office of your country, since the project itself says it cannot answer. Copyrighted works that the project hosts with the owner's permission follow other terms, written at the beginning of each such work. The table is a summary and drops details of the license. Commercial plans, translations and edited editions need the full text and, for the trademark, the project's own contact. We did not test mirrors, the harvest address or the feeds ourselves.

Our catalogs of free books

Five of our data catalogs cover Project Gutenberg books by genre. An entry gives the work's name and writer, its number in the collection, a download figure and an EPUB link, together with the date that link was last seen to answer. Each catalog flags its books as free in the US alone, so the rules above still apply to whoever uses them:

They save a person or an agent from searching the catalog for each title. They do not replace the license check, the rights check for your own country, or a visit to the landing page of the book.

Read this post as Markdown: /blog/public-domain-books-project-gutenberg.md · Atom feed.