An EPUB is a document that reflows, which is why page numbers do not survive.
Coming from the document side, the thing to know is that an EPUB has no fixed pages. The text reflows to whatever screen opens it, so there is no page 7 to cite and no layout to preserve. Converting to PDF or DOCX gives the text a page for the first time, and the converter has to invent where those pages break. Headings, lists, tables and links come across; fixed positioning never existed to lose.
Each chapter is a web page, the look comes from a stylesheet, and a manifest says what order to read them in. That's why EPUB text flows to any screen and why you can change the font. Open the folder of a short novel below.
<body> <h1>The Harbour</h1> <p>The ferry left at dawn. By the time the harbour lights went out behind them, …</p> <img src="images/harbour.jpg" alt="The harbour at dawn"/></body>Each chapter is a standard web page. No page numbers or positions, which is why the text can reflow.
An EPUB is a ZIP, so on its own it starts like any other archive. The container rules add one trick: a file called mimetype must come first, stored uncompressed, with no extra field. Then the words “application/epub+zip” always sit at byte 38, and software can recognise the book without unzipping it. Break a rule and watch the signature move or vanish.
Every file must appear in the manifest, but only the spine sets what comes next when you turn the page. A spine entry marked linear="no" is skipped by page turns and reached only through a link, which is how endnotes and pop-up answers usually work. Reorder the spine and see what a reader does.
<spine> <itemref idref="cover"/> <itemref idref="toc"/> <itemref idref="c1"/> <itemref idref="c2"/> <itemref idref="c3"/> <itemref idref="notes" linear="no"/></spine><!-- manifest lists all 6 files, plus style.css and images; nav.xhtml carries properties="nav" -->
EPUB Accessibility 1.1 asks publishers to describe a book’s features in the package metadata, using schema.org terms. Bookshops and reading apps turn these into the accessibility summary you see before buying. Since the European Accessibility Act took effect in June 2025, most new ebooks sold in the EU need this information. Tick what the book offers.
<meta property="schema:accessMode">textual</meta><meta property="schema:accessMode">visual</meta><meta property="schema:accessModeSufficient">textual</meta><meta property="schema:accessModeSufficient">textual,visual</meta><meta property="schema:accessibilityFeature">alternativeText</meta><meta property="schema:accessibilityFeature">tableOfContents</meta><meta property="schema:accessibilityFeature">structuralNavigation</meta><meta property="schema:accessibilityHazard">none</meta><meta property="schema:accessibilitySummary"> All images are described. Headings follow the chapter structure.</meta>
An EPUB works when every reference inside it resolves. That means every link, image, stylesheet and font, plus the manifest and the navigation. This page parses each XHTML page and stylesheet with your browser’s XML parser and resolves every reference against the package. It checks fragment targets against the ids that really exist, compares each declared media type with the file’s first bytes, and works out which manifest properties each page needs. Then it writes a repaired copy and checks it again. Readers differ in how forgiving they are: a broken reference that one app ignores can stop another, or fail a shop’s upload check. Everything happens on your device.
Most readers open all three. The differences show up in accessibility, audio and page design.
Treating an EPUB as a document runs into one structural fact immediately: there are no pages in the file. Each chapter is a standard web page with a stylesheet, and the manifest lists what is inside while the spine decides the reading order. Nothing anywhere records where page 7 begins, because the text is meant to reflow to whatever screen opens it. Change the font size on a reader and every boundary moves.
That is why a reference to an EPUB cannot be a page number. Readers invent their own locations instead, and those locations differ between devices and sometimes between app versions on the same device, so two people reading the same file will not agree on where a passage sits. If you need a stable citation, you need a fixed-page format, which means converting to PDF and then citing the PDF rather than the book.
Converting the other way, into DOCX or ODT, gives the text pages for the first time. The converter has to decide where those pages break, because the source never said: it fills a page, starts another, and keeps going. Headings, lists, tables, bold and links survive the trip because they are ordinary HTML in the source and have direct equivalents in a word processor. What cannot survive is fixed positioning, since the EPUB never had any to lose.
Images are the part worth checking afterwards. In the book they are sized relative to the screen, so a picture declared at ninety per cent width adapts to the reader. In a paginated document that same picture needs an absolute size in millimetres or inches, and the converter picks one. Wide tables have the same problem in sharper form: a table that scrolled sideways on a tablet has to fit a page or be broken across several.
If the goal is simply to read the book on a computer rather than to edit it, converting may be unnecessary. EPUB opens in several desktop readers and in the browser, and the text stays reflowable, which is the format working as designed. Convert when something downstream genuinely requires pages: a print workflow, a submission form that takes DOCX, or a document that has to be marked up and returned.
Identifiers and rules from the W3C EPUB 3.3 Recommendation, which now defines the format, and the steps from Open eBook to today.
Each page explains what's kept and what changes.