The short version

A PDF holds a page. An EPUB holds a book, and lets the device decide what a page is. The conversion throws the layout away deliberately, which is the only way text can reflow to a phone, an e-reader or somebody's preferred type size.

Headings are worked out from how the text is set, the book is split into a chapter per heading, and the chapter list is shown before anything is written.

Most people asking for this have the same problem: a document they want to read, on a device that makes a PDF miserable. Everything below follows from that.

Fixed and reflowable are opposites

A PDF describes a sheet of paper. Every line is at a coordinate, in a font at a size, and it will be there whatever you open it with. That is exactly what you want for a contract and exactly what you do not want on a phone, because the only way to fit an A4 page on a five inch screen is to make the text too small to read.

The same documentWhy a page and a screen disagree
A PDF on a phoneThe page keeps its size. You pan.An EPUB on the same phoneThe text fits the screen. You read.
An EPUB carries the text and the structure and no geometry at all. The reader lays it out at the width it has and the type size somebody chose, which is why a book reflows and a PDF does not.
What you gain

Text that fits any screen at any type size.

Line length and font chosen by the reader, not by the document.

Search, highlights, notes and a working table of contents.

A file a tenth the size, because there is no page to describe.

Works with a screen reader, because the structure is real.

What you give up

The layout. Columns, boxes, pull quotes and placement are gone.

Page numbers, which an EPUB has no concept of.

Images, in this version.

Anything that depended on where it sat on the paper.

Exact typography. A reader picks the font, not you.

This is the trade, not a shortcoming
Any tool that claims to convert a PDF to an EPUB with the layout intact is describing a fixed-layout EPUB, which is a picture of each page wrapped in an e-book, and which reflows nowhere. If the layout matters, keep the PDF.

Chapters, and why they matter more than you would think

An EPUB reader paginates one file at a time. A whole book in a single chapter means it lays out the entire text on every page turn, which on a phone is a stutter and on an e-ink reader is a second and a half of nothing happening.

What the file holdsThe package, and the reason for the split
Inside the .epub, which is a zipmimetypefirst, and uncompressedMETA-INF/container.xmlpoints at the packageOEBPS/content.opfmanifest and reading orderOEBPS/nav.xhtmlcontents, EPUB 3OEBPS/toc.ncxcontents, older readersOEBPS/ch001.xhtml …a file per chapterWhy a chapter is a fileA reader paginates one file ata time. A whole book in onechapter means it repaginatesthe lot on every turn, whichon an e-ink device is thedifference between instantand a second and a half.So the split is shown beforethe book is written.
The mimetype entry has to be first in the zip and stored rather than compressed, which is the one rule the format makes about its own container. Everything after that is the book.
Written into every bookWhat the package carries
  • EPUB 3, with a navigation documentmodern readers
  • And the EPUB 2 table of contentsolder devices
  • One XHTML file per chapterso pagination is quick
  • A stylesheet that gets out of the waythe reader decides

Chapters are found by heading, and a heading is found by being larger or bolder than the body text around it. A book that sets its chapter titles properly splits cleanly. One that does not gives no signal at all, and the chapter list will say so by showing a single entry.

Split
At headings

First level, or deeper if the book needs it.

Contents
Both kinds

EPUB 3 nav and the older NCX.

Chapters
Well formed XML

A reader rejects a book over one unclosed tag.

Which PDFs convert well

The honest answer is that it depends almost entirely on how the PDF was made, and you can predict it before you try.

1

A book, report or manual set in one column

Converts well. Headings are usually set larger, paragraphs are ordinary, and the result reads the way you would hope.

2

A thesis, a white paper, a manuscript

Converts well, with the same caveat about headings. Footnotes at the foot of each page arrive as paragraphs in the flow.

3

A two-column academic paper

Converts badly, and predictably so. See below.

4

A magazine, a brochure, a form, a slide deck

Do not. These are layout, and layout is the thing being discarded.

The case that goes wrongTwo columns, one baseline
Two columns on the pageWhat a baseline seesOne line, read straight across.
Lines are grouped by where they sit on the page, so text in two columns at the same height is read as one line running across both. The result is searchable and not readable, and no amount of tuning fixes it without column detection, which this does not do.

Saying that plainly is more use than a tool that produces the mess and lets you find out. If a paper is what you have, converting it to Markdown and fixing the order by hand is often faster than fighting it.

Getting it onto a reader

Every reading device and app in current use takes EPUB. Kobo, Nook, Apple Books, Google Play Books, Calibre and every phone app read it directly, usually by dropping the file in over USB or emailing it to yourself.

Kindle is the one that used to be awkward and no longer is: Send to Kindle takes EPUB, by email to your device address or through the app, and Amazon converts it at their end. Sending a PDF the same way gets you a PDF on the device, which brings you back to the problem you started with.

What does not come through

  • Images. The text, headings, lists and tables convert; pictures do not. Pull them out separately if you need them.
  • Page numbers and cross-references to them. An EPUB has no pages, so "see page 42" is a sentence about a document that no longer exists.
  • Running headers and footers. Dropped on purpose, because a line repeating every page is furniture rather than text. The count of what went is in the result.
  • Links and bookmarks from the PDF. The structure is rebuilt from the layout rather than carried over.
  • Anything in a scan. A picture of a page has no text in it. OCR first.

Why this runs in your browser

A book somebody is converting is usually one they bought, wrote, or were sent in confidence. None of those improve by being uploaded to a converter, and the usual excuse, that this kind of work needs a server, stopped being true some years ago.

The PDF is read in memory with the same engine your browser already uses to display one, the structure is worked out, the package is zipped and handed back. Nothing is transmitted and nothing is stored.

The tool is right above

Drop a PDF, check the chapter list, and take the book.

Jump to the tool

Common questions

The trade being made, the documents that do not suit it, and getting the file onto a device.

01Will the EPUB look like the PDF?

No, and it is not meant to. An EPUB has no pages of its own: the reader makes them, at whatever size and typeface the person reading has chosen. Giving up the original layout is the entire point of converting, and a tool that promised to keep it would be promising a PDF with a different extension.

02Does it work on a scanned book?

No. A scan is photographs of pages with no text in them, and an EPUB is text. There is nothing to put in the file, so it says so rather than handing you an empty book. Run the scan through OCR first and then convert.

03Are images included?

Not yet. The text, the headings, the lists and the tables come through; pictures do not. For an illustrated book that matters, and for the novels, reports, papers and manuals most people are converting it usually does not.

04Why did the whole book come out as one chapter?

Because nothing in the PDF marks a chapter. Headings are recognised by being larger or bolder than the body text, and a document that sets its chapter titles at the same size as its paragraphs gives no signal at all. Try splitting at the second level, and if that does not help the PDF genuinely has no structure to find.

05How do I get it onto a Kindle?

Send to Kindle takes EPUB directly now, by email or through the app, and converts it at Amazon's end. Kobo, Nook, Apple Books and every reading app on a phone take EPUB as it is.

06What about a two-column academic paper?

Poorly, and this is worth knowing before you try. Lines are grouped by where they sit on the page, so two columns at the same height are read as one line running across both. A paper converts to something you can search and not something you would want to read. Print the columns to a single column first if the PDF came from a tool that can.

07Is the file a real EPUB?

It is EPUB 3, with the navigation document a modern reader wants and the older EPUB 2 table of contents that a good many devices still in daily use look for first. The mimetype entry is stored uncompressed and first in the zip, which is the one rule the format makes about its own container, and every chapter is well formed XML, because a reader rejects a whole book over a single unclosed tag.

08Is anything uploaded?

No. The PDF is read, the structure worked out and the book written in your browser. Nothing is transmitted and nothing is stored.