Calibre PDF to EPUB: how to convert, and where it struggles

The settings that change the result

The defaults are a sensible start. If the EPUB comes out messy, these are the options in the Convert books window that make a difference with PDFs:

• PDF input, line un-wrapping factor (0.45 by default): how short a line has to be before Calibre treats it as the end of a paragraph. Lower it if paragraphs are cut in the middle, raise it if separate paragraphs get glued together.
• PDF input, PDF engine: the newer "calibre" engine tries to remove headers and footers by itself; the older pdftohtml engine doesn't. The calibre engine also takes a regular expression for the first or last line of each page, for a running head that keeps slipping through.
• Heuristic processing (off by default): extra passes that rejoin broken lines, rebuild chapter headings and delete blank paragraphs. Worth one try on a messy book; check the result, since it can also join lines that should stay apart.
• Structure detection and Table of contents: where Calibre looks for chapter starts. If the contents list is empty or full of junk, this is the place to adjust.

Calibre has no OCR. A scanned PDF without a hidden text layer comes out as pictures of the pages, which you can view but not search or resize, and in our test a scan that did have a text layer came out the same way. Calibre's own manual is frank that PDF is one of the hardest formats to convert from, because a PDF stores where each line sits on the page rather than how the text flows.

What we saw when we measured Calibre's EPUBs

We ran Calibre 9.13 with its default settings on five public-domain PDFs and scored every EPUB with the same script as seven other converters. Some of what it left behind:

• Psychology of Intelligence Analysis (CIA): 62 stray page numbers ended up in the reading text, 373 paragraphs stopped mid-sentence, and every "ff", "fi" and "Th" came out as one letter ("efort", "frst", "Te"), because the PDF's character map gives those letter pairs wrongly.
• IRS Publication 501, set in three narrow columns: 269 words stayed split at a line-end hyphen and 347 paragraphs were cut short.
• A scanned chapter with no text layer: no text at all, only page pictures.
• The 300-page 1876 Tom Sawyer scan: more than three hours on a laptop that was also busy with other work, and the result was an 824 MB EPUB of page pictures, with none of the scan's text.

The hardest file was the 1882 Scientific American, whose columns sit a few points apart. Calibre kept 89.0% of its words to our 95.9%, and mixed lines from neighbouring columns on most pages. Looking to replace Calibre altogether? See Calibre alternatives by job. The full tables, with every tool and every file, are in our PDF to EPUB converter comparison.

When to use Calibre, and when to use epubwise

Use Calibre when the file must stay on your computer, when you are converting a whole folder at once, or when the PDF is plain running text from a word processor. It is also a good free tool for organising a library and sending books to an e-reader over USB.

Use epubwise when the PDF fights back: a scanned book that needs OCR, pages with two or three columns, running heads on every page, or tables. It is built for those cases, as the step-by-step explanation shows. You can also do both: convert the PDF here, then add the EPUB to Calibre like any other book.