Convert PDF to EPUB

A PDF has pages. An ebook does not — and that is the whole idea

Everything this page does follows from one difference between the two formats. A PDF records where every line sits on a sheet of a fixed size, because it was designed to print the same way on any machine. An ebook records the words and lets the reading device decide where the lines fall, because it was designed to be read on a phone, a six-inch e-reader and a tablet without being redrawn for each.

That means the page breaks in your PDF have no counterpart in the file this tool makes, and removing them is not a compromise — it is the conversion. Keeping them would mean inserting a paragraph break roughly every forty lines of the finished book, which looks almost passable on a laptop and falls apart on the small screen the format exists to serve. Sentences that ran across a break are stitched back together, hyphens at the seam included. If your document genuinely needs its fixed layout — a form, a score, a diagram-heavy manual — the honest advice is to keep the PDF and read it as one.

PDF and EPUB, in brief

PDF

Full name	Portable Document Format
Extension	.pdf
Format type	Fixed-layout document
MIME type	application/pdf

EPUB

Full name	Electronic Publication
Extension	.epub
Format type	A reflowable ebook — a zip of XHTML with no fixed pages
MIME type	application/epub+zip

Nothing works unless the characters are actually in the file

A PDF does not store text the way a document does. It stores instructions for drawing shapes at coordinates, and whether those shapes come with the underlying characters attached depends entirely on how the file was made. A PDF exported from a word processor carries them. A PDF made by photographing or scanning paper does not: it holds pictures of letters, and the letters themselves were never written down anywhere in it.

This matters more here than almost anywhere else, because a scan and an export look identical on screen. Feed a scan to a converter and it will work away happily and hand back an ebook whose chapters are blank — and you find out after the download, not before. So the check runs first and the verdict appears before there is anything to click: every page has text, some pages are pictures, or there is no text at all. In that last case no file is offered, because offering one would be offering an empty book. What that document needs is optical character recognition, which examines the shapes and guesses at the letters. It is a genuinely different operation, it makes mistakes a conversion cannot make, and no tool that promises a straight conversion is doing it.

If you only want the words rather than a book, extracting the text from a PDF reports the same thing page by page and hands back a plain file.

Running heads and page numbers are removed — but only when they repeat

A printed book puts its title along the top of every page and its number along the bottom. Inside a PDF those are ordinary text at ordinary coordinates, indistinguishable from a sentence. Inside an ebook they belong nowhere at all: there is no page for them to head, so each one lands as a stray line dropped into the middle of a paragraph, over and over, for the length of the book.

The tempting fix is to delete the first and last line of every page, and it is wrong. Measured against a thousand-page act of legislation, the running head appeared at the top of 149 pages out of 150 — while the bottom line never repeated at all, because it was simply wherever the prose happened to reach. One page ended on the fragment gram., the tail of a word that had been split as pro- at the foot of the previous one. Stripping by position would have swallowed real text on every page in the file, silently, and the result would have read almost normally.

So repetition is the only evidence used. Lines at the top and bottom of each page are compared with digits masked out, which makes Page 7 of 300 and Page 8 of 300 the same line, and something is removed only when it turns up across many pages. A book with no running heads loses nothing, and the receipt names the pattern that was taken out so you can see whether the judgement was right. Pages with only a handful of lines are left alone entirely — on a title page there is no way to tell a heading from the content.

What a PDF to EPUB conversion cannot carry across

Tables come across as their text, in reading order, without the grid. That is not laziness; a reflowing format has no fixed columns to hold a table in, which is why publishers who care about tables ship them as images. Mathematics fares worse — an equation is positioned glyph by glyph, and flattening it gives you the symbols in an order nobody can read. Illustrations do not come across in this version at all. Pulling them out means decoding embedded objects stored in several unrelated compression schemes, and rather than half-implement that and drop some pictures without saying which, this tool takes the text and tells you plainly that is what it took.

Chapters are found by looking at type size and at wording — a line set larger than the body around it, or one that opens with a word like Chapter or Appendix. It is a judgement rather than a fact, since a PDF records no structure to read, so the count is shown on the receipt and the chapter list is there to be checked. Where no headings turn up, the text is divided into parts of about twenty pages and labeled as ours rather than the book's.

The title and author come from the file, or from you

Every ebook needs a title, because that is what your library shelf displays. PDFs carry a title field, but it is almost never filled in by a person — it is set by whichever program wrote the file, so it commonly reads Microsoft Word - report_final_v3.docx, or untitled, or the name of a slide. Copying that straight into your library is something a tool does when it has forwarded the metadata rather than read it.

Those patterns are recognized and flagged instead, along with author fields holding a computer account name rather than a person. The boxes are yours to correct before the book is built, and where the file said nothing, the box is left empty rather than filled with a guess. The one field that cannot be left blank is the language, since reading systems use it for hyphenation and for speech, so a default is filled in and labeled as a default. Should you want the words back on paper afterwards, the conversion in the opposite direction sets them again on a page size you choose, and Markdown to PDF uses the same typesetting for notes that never were a book.

The job people want done on a device they gave back

The established answer here is Calibre, and it is a fine program — a whole library manager, two decades of work behind it, free. It is also a five-minute setup of something you may want exactly once. The document tends to be a course reader, a dissertation, a service manual or a public-domain scan, collected somewhere that installing is simply not among the options: a study-room terminal, a laptop loaned until the end of term, a computer in an archive's reading room, a family desktop nobody wants another program on.

Sessions like those end cleanly by design, which is both the appeal and the catch — sign out of a shared login and whatever you arranged goes with it. Working in a tab suits that shape. Nothing is added to the machine, the finished book drops into the downloads folder, and the sensible next move is to put it on the e-reader or the phone before you stand up. Everything in between lives in memory for as long as the tab does, so what survives the visit is the one file you asked for.

The other files that turn into an EPUB

A PDF is the hard case, because the characters have to be recovered from drawing instructions and might not be there at all. Two other formats are the easy case, and they are worth knowing about if the shelf you are tidying has more than PDFs on it. An old Kindle .mobi already holds its text as markup, so converting MOBI to EPUB is a matter of unpacking rather than reconstruction — there is no scan to worry about and no running heads to detect, because a book that was already an ebook never had pages for them to sit on.

AZW3 to EPUB does the same for the newer Kindle format, and it ends up somewhere this page cannot reach: because an .azw3 keeps its chapters as separate documents, the division comes back exactly as the publisher set it, titles included. Chapters here are a judgement — type size and wording, checked on the receipt because a PDF records no structure to read — whereas there it is a fact recorded in the file. The one thing neither of those pages will do is open a book still locked to a Kindle account.

Sending it to a Kindle, and where all this happens

Send the .epub itself. Amazon stopped accepting MOBI through Send to Kindle at the end of 2023 and takes EPUB directly, so converting again would only cost you the reflowing that makes a small screen bearable in the first place — which is precisely what the PDF you started with could not do.

The file is opened, its text layer read, and the ebook archive assembled by code running in the tab you have open. Rather than ask you to believe that, check it: open your browser's Network panel, drop the document in, and watch the request list stay empty of it. One request does appear the first time you hand a document over, and it is worth naming rather than glossing: the page collecting its own reader, from this domain, once. Nothing of yours goes with it, and afterwards the network takes no further part in either direction. Services that do this on a server say much the same thing in gentler words — one of the better-known ones notes that files are deleted after a few hours, which is a fair description of a book sitting on someone else's computer for a few hours.