Where to find well-made public domain ebooks

Public domain books are free, which makes it easy to forget that the edition still matters enormously. Two copies of the same novel can differ by decades of editorial work.

The sources worth using

Standard Ebooks — the best of them by a wide margin. Volunteers take public domain texts and typeset them properly: modernised punctuation, real semantic markup, considered covers, valid EPUB 3. Small catalogue, uniformly excellent.

Project Gutenberg — the largest and oldest archive, over 70,000 titles. Quality varies because the texts span decades of transcription practice. The EPUBs are functional rather than beautiful.

The Internet Archive — enormous, and much of it is raw scans. Useful when nothing else has the book; expect to do work.

Wikisource — collaboratively proofread against page scans, so accuracy is good. Export quality is inconsistent.

Telling a good edition from a bad one

Other tells: OCR artefacts (rn where m should be, 1 for l), page numbers and running headers stranded mid-paragraph, and no metadata — a book that shows as “Unknown Author” was not made with care.

Fixing what you have

If the only available edition is rough, most of it is repairable. Run it through the Validator to find structural problems, fix the title and author with the Metadata Editor, and re-typeset it in Style Studio — a decent typeface and a sane line length transform a plain Gutenberg text.

OCR errors in the text itself are the exception. Nothing can fix those but a better source, which is the strongest argument for starting at Standard Ebooks when they have the title.

What “public domain” actually covers

Worth being precise, because the edition matters as much as the work.

In most jurisdictions copyright expires a set number of years after the author’s death — 70 in the EU, UK and US for most modern works. Once expired, the text is free for anyone to use.

But a specific edition can carry its own rights. A new translation is a new copyrighted work. So is a scholarly edition with original annotations, and often the typesetting and cover design. Dickens is public domain; a 2019 annotated Dickens with a new introduction is not, in those parts.

For reading this rarely matters. For republishing or building on a text, check which edition you started from.

Comparing the sources

The practical rule: check Standard Ebooks first. If they have it, stop looking — the edition will be better than anything you would produce yourself. If not, Gutenberg. Fall back to the Archive only when nothing else has it.

What careful production looks like

The difference between a good and bad free edition is not the words, it is everything around them:

• Real semantic markup — headings that are headings, so navigation works
• A working table of contents
• Modernised, consistent punctuation — curly quotes, proper em dashes, no OCR artefacts
• Metadata that is filled in — title, author, language
• A cover — even a plain typographic one
• Italics preserved from the original, not flattened

Standard Ebooks does all of this by policy. Gutenberg’s older texts often do none of it, because they were transcribed when the goal was preserving the words at all.

Spotting OCR damage

Scanned texts carry characteristic errors, and once you know them you cannot unsee them:

• rn read as m — “modern” becoming “modem”
• l / 1 / I swapped
• cl read as d
• Long-s (ſ) in older printings read as f — “ſhall” becoming “fhall”
• Running headers and page numbers stranded mid-paragraph
• Hyphens left in from line breaks: “conver- sation”

Search the file for “modem” and “fhall”. If either appears in a book that has nothing to do with modems, the whole text was OCR’d without proofreading, and there will be hundreds more errors you have not found.

Making a rough edition better

Structural problems are fixable in minutes; textual ones are not.

Run it through the Validator to find what is actually broken. Fix the title and author with the Metadata Editor so it files correctly in your library. Then open it in Style Studio — a decent typeface, a 60–75 character measure and sane leading transform a plain Gutenberg text into something genuinely pleasant.

None of that repairs OCR errors in the words themselves. Those need a better source, which is the strongest argument for starting at Standard Ebooks whenever they have the title.

Giving something back

All of these are volunteer projects. Gutenberg’s Distributed Proofreaders lets you proofread a page or two at a time against the original scan; Standard Ebooks takes contributors for typography and production. If you read a lot of free books, an hour occasionally is a fair trade.