What Is a PDF File? Portable Document Format, and What Is Inside One
A PDF file is a page-for-page snapshot of a document, frozen so it lands the same way everywhere. Jump to your question below, or read it through in order.
A PDF file is a page-for-page snapshot of a document, frozen so it lands the same way everywhere. Jump to your question below, or read it through in order.
PDF stands for Portable Document Format. A PDF file holds text, fonts, images and the exact position of every one of them. So pages land the same way on any computer that opens them.
Adobe built it in the early 1990s. Since 2008 it has been an ISO standard, and a committee rather than Adobe decides what goes into it.
Send somebody your Word file and you are sending instructions. Their computer rebuilds the document from fonts and margins it happens to have, and the result drifts.
Send PDF instead and you are sending the finished page. Every letter already has coordinates. Nothing gets recalculated at the other end, which is why one contract printed in Mumbai matches the same contract printed in Munich.
Independence from the machine is the entire point. PDF presents documents the same way whatever the software, hardware or operating system at the other end.
That single design choice explains almost everything else on this page. PDF is hard to edit because it stores positions, not paragraphs. It is reliable for the same reason.
.pdfapplication/pdf, registered in RFC 8118%PDF- then the version number%%EOFSearch this format and you will be told it means Portable Document File. That version is everywhere, including on this page until today.
It is Portable Document Format. ISO 32000 is titled Document management, Portable Document Format. Adobe, who named the thing, put it plainly.
“PDF is an abbreviation that stands for Portable Document Format.”
Worth saying plainly because the wrong version travels. Quote it in your specification, tender or compliance document and somebody will eventually notice.
PDF grew out of PostScript, the printing language Adobe already sold. PostScript described pages to printers. PDF describes pages to screens, with the parts printers do not need stripped out and the parts readers do need added.
You can see the family resemblance in the PostScript file format, and in EPS graphics files, which came from the same place.
Every PDF has four parts in the same order. Header, body, cross-reference table, trailer. The table records where every object sits, and the trailer closes with %%EOF.
Open any PDF in Notepad and the first line is readable. After that it turns to noise, because most of the file is compressed. Underneath, the shape is simple.
%PDF- and the version number. RFC 8118 puts it plainly, that all PDF files start with those characters followed by the version.xref. For every object it records the byte offset from the start of the file, which is how readers jump straight to page 800 without touching pages 1 to 799.startxref says where the index begins. Then the root object, then %%EOF. Readers open PDF files here and work backwards.Part three is the one to remember. That index is what makes 900-page reports open instantly, and it is what breaks when the file arrives incomplete.
Readers look at the end of the file for that index, then follow byte offsets to reach anything else. Lose the last few kilobytes and every offset still points at bytes that are no longer there.
That is why half-downloaded PDF files are dead rather than partly readable. Truncate a JPEG and you still see the top half of the picture. Truncate a PDF and you see nothing, because the map to the contents lived in the part that never arrived.
Section 11 covers what to do about it, and the answer is usually simpler than people expect.
Adobe published the PDF specification from 1993 to 2007. In 2008 ISO took it over as ISO 32000. The current edition is ISO 32000-2:2020, and since 2023 it has been free to download.
Most pages on this subject still describe PDF as an Adobe format. It was one for fifteen years. It has not been one for eighteen.
It started in 1991 as an internal Adobe proposal from co-founder John Warnock, called the Camelot Project. The goal was one file anybody could view and print, whatever machine they had in front of them.
Adobe gave the format away early, which is the reason it won. The specification was free and royalty-free from the start, so anybody could write software that produced or displayed PDF without paying a licence.
That was unusual. Plenty of file formats stayed locked to one vendor and died with the product that made them.
“In 2008 ISO published Adobe’s PDF Reference 1.7 (with minor, non-technical changes) as ISO 32000-1.”
So ISO 32000-1 is PDF 1.7 with the Adobe branding taken off. Nothing technical changed at the handover. What changed is who decides what happens next.
The International Organization for Standardization, ISO for short, publishes it now. PDF has been an open standard rather than one company's property ever since.
PDF 2.0 is the proof. It arrived in 2017 from committee work, with no Acrobat launch beside it. It also dropped Adobe's own XFA forms and JavaScript extensions from the standard.
Ignore any page that says the format changes when Adobe ships a new Acrobat. That was true up to PDF 1.7 in 2006. It has been wrong for every version since, and it leads people to assume Adobe can still change the rules.
ISO TC 171 SC 2 maintains the standard. The PDF Association runs the secretariat.
ISO standards normally cost money, and for years ISO 32000-2 did. In April 2023 the PDF Association, with Adobe, Apryse and Foxit, made it available at no cost.
Almost nothing written about PDF mentions this. If you are building software, arguing a compliance point or settling a bet, the actual text of the standard is a click away.
Nine versions, from PDF 1.0 in 1993 to PDF 2.0 in 2017. Your file names its own version on the first line, so a text editor tells you in two seconds.
Version numbers matter less here than in most formats, because readers stay compatible in both directions. They still explain why old files behave oddly in new tools.
| Version | Year | Shipped with | Headline addition |
|---|---|---|---|
| PDF 1.0 | 1993 | Acrobat 1.0 | The format itself |
| PDF 1.1 | 1994 | Acrobat 2.0 | External links and basic security |
| PDF 1.2 | 1996 | Acrobat 3.0 | Forms, and Flate compression |
| PDF 1.3 | 1999 | Acrobat 4.0 | Digital signatures |
| PDF 1.4 | 2001 | Acrobat 5.0 | Transparency |
| PDF 1.5 | 2003 | Acrobat 6.0 | Layers and object streams |
| PDF 1.6 | 2005 | Acrobat 7.0 | Embedded 3D |
| PDF 1.7 | 2006 | Acrobat 8.0 | Later became ISO 32000-1 |
| PDF 2.0 | 2017 | Nothing | ISO committee release, no Adobe launch |
Read the last row again. PDF 2.0 is the first version of this format that was never an Adobe product launch, and the only one with an empty column there.
Use this when a tool rejects your file as unsupported, or a client asks for PDF 1.4 and you have no idea what you sent them.
%PDF- and then a number, so %PDF-1.4 or %PDF-2.0.Old and new readers cope with each other better than you would expect. A PDF 1.0 file opens in today's software, and a PDF 2.0 file opens in software from 2005. Older software ignores features it was never taught, so pages may lose transparency or 3D objects and still read fine.
One catch. The version on line one is what the file claims. Tools that add features later can leave that line alone, so treat it as a strong hint rather than a guarantee.
The real types of PDF are ISO subsets: PDF/A for archiving, PDF/X for print, PDF/UA for accessibility, PDF/E for engineering and PDF/VT for variable print. They restrict PDF rather than extend it.
Ask what types of PDF exist and the answers come back muddled. Here is the honest one. ISO defines five subsets, and every one of them is ordinary PDF with rules bolted on.
| Subset | Standard | Built for | You meet it when |
|---|---|---|---|
| PDF/A | ISO 19005 | Long-term archiving | A court, a bank or a records office asks for it |
| PDF/X | ISO 15930 | Prepress and printing | A printer sends your artwork back |
| PDF/UA | ISO 14289 | Accessibility | A public body has to meet access rules |
| PDF/E | ISO 24517 | Engineering drawings | Construction or manufacturing documents |
| PDF/VT | ISO 16612-2 | Variable data printing | Personalized statements and mailers |
Notice what they have in common. Every one of them forbids things. Fonts must be embedded. Colors must be device-independent. Certain compression filters are off the menu.
PDF/A goes furthest. No JavaScript, no audio or video, no links to files that might vanish. The point is a document that still renders correctly in thirty years, when nothing it depended on is around.
PDF/X changed printing the same way. Studios once sent encapsulated PostScript artwork to a printer file by file. One PDF/X carries the entire job instead, fonts included.
These four are not types of PDF: PDF Formatted Text and Graphics, PDF Searchable Image Exact, PDF Image Only, PDF Searchable Image Compact.
They are scan output settings from one scanning application called PaperPort. They were sitting on this very page as the official list until today. If a page offers you those four, it copied them from the same place we did.
There is a genuine distinction hiding behind that list, and it is the one people actually mean when they ask about types. Section 7 covers it: a PDF made from a document and a PDF made from a scanner are very different animals.
PDF looks the same everywhere only when the fonts travelled with it. Embedding is optional in ordinary PDF, so files built without it borrow substitutes from whatever computer opens them.
Every page about this format makes the same promise. The same PDF file looks identical on every machine that opens it. Mostly true, and the part that gets left out is worth knowing.
PDF can carry its fonts inside the file. It does not have to. When it does not, it names the font it wants and leaves your computer to find something close.
“Faithful rendering requires that fonts be embedded.”
Borrowing fonts rarely ruins pages. It shifts them. Letters get wider or narrower, lines wrap in new places and the table that fitted on one page now runs onto two.
That is the entire reason PDF/A, PDF/X and PDF/E all make embedding compulsory. Somebody noticed the promise had a hole and wrote standards around it.
The file carries its own type. Layout is identical on machines that have never seen that typeface. Larger file, and the only version you can rely on.
The file names its fonts and hopes. Smaller, fine inside one company where everybody has the same fonts installed, risky the moment it leaves.
You can check in about ten seconds. Open File, then Properties, then the Fonts tab. Entries marked Embedded or Embedded Subset came with the file. Anything without that word is being borrowed from your machine.
Worth doing before sending artwork to a printer, and before archiving anything you expect to still open in ten years.
Two kinds of PDF arrive in your inbox. One was exported from your word processor and holds real text. One came off the scanner and holds a photograph of text. Try to select any word and you will know in seconds.
Both files end in .pdf. Both open in the same reader. What you can do with them afterwards is completely different, and nothing on the outside of the file tells you which you have.
Drag your cursor across one line of the document. If words highlight, there is real text in there. If a blue rectangle covers the area instead, you are looking at a picture.
That third row is the one worth knowing about. OCR does not replace your scan. It adds a hidden layer of characters positioned on top of the picture, so search and copy have something to grab.
It is also why scanned pages sometimes search perfectly and sometimes return nonsense. OCR guesses, and faint faxes or handwritten notes give it very little to work with.
A scanned PDF that has never been through OCR is invisible to search engines and to your own computer's search too. Archive ten years of paperwork that way and you have a filing cabinet, not a database.
Most scanning software offers OCR as a checkbox at scan time. Ticking it costs seconds per page and saves you rescanning everything later.
Neither kind is better. Scans are the legal record of a signed page. Exported PDF is the working document. Trouble starts when somebody needs one and was handed the other.
PDF has two passwords and they are not equal. A user password encrypts the file. An owner password only sets flags that readers are asked to respect, and free tools remove it in seconds.
Protect any PDF and your software offers two boxes. They sit next to one another, they look equally serious and one of them is not security at all.
| Password | Also called | What it does | How strong |
|---|---|---|---|
| User password | Open password, document open password | Encrypts the file. Without it there is nothing to read | Real. Modern PDF uses AES |
| Owner password | Permissions password, master password | Sets flags saying no printing, no copying, no editing | A request. Stripped in seconds |
The user password does what you expect. Content is encrypted, and somebody without the password has nothing but noise.
The owner password does not encrypt anything. It writes a note into the file asking readers to disable printing or copying, and well-behaved software honours it.
An owner password protects nothing from anybody who does not want to cooperate. The content is sitting there unencrypted. Software that ignores the flag reads it straight out, and plenty of free tools do exactly that.
Competing guides describe this setting as locking your PDF so it cannot be edited or altered. It does not lock anything.
Use the owner password for what it is good at, stopping an honest colleague printing your draft by accident. For anything that actually matters, set a user password or do not send the file.
Rule of thumb. If somebody stealing the content would be a problem, encryption is the only answer. Permission flags are a courtesy notice, not a lock.
Digital signatures are a third thing entirely, and often the one people actually want. They do not stop anybody reading the document. They prove it has not changed since it was signed.
Images are almost always the reason a PDF is large. Text compresses to nearly nothing, while a page scanned at 600 dpi is a photograph. Repeated saving adds to it, because a change can be appended rather than rewritten.
One forty-page contract can be 300 KB or 90 MB. Same number of pages, same words. The difference is what the file carries besides the words.
PDF does not use one compression method. It stores a name against every stream so readers know how to unpack it.
| Filter | Used for | Notes |
|---|---|---|
| FlateDecode | Text and most page content | The same Deflate used by ZIP. Lossless, and the common one |
| DCTDecode | Photographs | JPEG. Lossy, so quality drops every time it is recompressed |
| JPXDecode | High quality images | JPEG 2000. Banned in several of the subset standards |
| LZWDecode | Older files | Superseded by Flate. Still turns up in files from the 1990s |
Use this when a file bounces off an email size limit, or an upload form rejects it.
There is a second kind of slow, and shrinking the file does not fix it. A PDF meant for the web can be linearized, also labelled Fast Web View. It moves everything page one needs to the front, so a browser shows that page before the download finishes.
Without it a browser waits for the last byte, because the index it needs lives at the end. Same file size, very different wait.
Everything opens PDF. Every browser renders it, and Windows, macOS, Linux, Android and iOS all ship a reader. You need Adobe Acrobat Reader for PDF about as much as you need it for JPEG.
This was the hard part in 1995 and it stopped being a question years ago. Software that cannot display PDF is now the unusual case.
Two claims on the old version of this page were simply false, and both are repeated across the web. PDF is not unsupported on Linux, and Acrobat Reader is not paid.
Reader is free and always has been. Acrobat Pro is the paid product, and it is an editor. Confusing the two sells people software they do not need.
Creating a PDF is just as ordinary. Word, Excel, Google Docs, LibreOffice and macOS all export one from a Print dialog. Vector tools go further and treat PDF as a native container.
That is why a CorelDRAW vector drawing and a Photoshop document both export a PDF that a printer accepts without conversion.
An Adobe Illustrator AI file has been a PDF underneath since Illustrator 9, which is why some of them open straight in a PDF reader. Photoshop PSD files can be saved the same way.
A PDF that will not open is usually a download that stopped early. Because a reader starts at the end of the file and works backwards, a missing tail leaves it with nothing to follow. Download it again first.
Readers are vague about this on purpose. The file is damaged or could not be repaired covers a dozen different faults, and the wording gives you nothing to act on.
People call these corrupt PDF files. Usually nothing is corrupt. Bytes are missing, and the reader has no way to say so politely.
| What you see | What it usually means | Start with |
|---|---|---|
| Damaged or could not be repaired | The file is missing its tail, so the index is gone | Download it again |
| Blank or grey pages | Page content is there, fonts or images are not | A different reader |
| A password box you did not expect | A user password was set on it | Ask the sender |
| Opens, but every character is garbage | Broken or missing font encoding | Ask for a re-export |
| Not a PDF at all | Something renamed, or an email attachment that arrived wrong | Check line one in a text editor |
Every step here is free, and the first two settle the majority. Repair software belongs at the bottom, not the top.
%PDF- on line one. Renamed files give themselves away here.No repair tool can recover data that never reached your computer. Pages advertising recovery for a truncated download are selling you a scan of the bytes you already have.
Work out whether the file is incomplete or merely malformed before paying for anything. The size check in step 1 answers that in ten seconds.
One more cause worth naming. PDF can carry attached files and JavaScript, and RFC 8118 warns that readers must run any of it in a protected environment.
Modern readers sandbox by default and often block scripts outright. So any PDF asking for permission to run something, from a sender you do not recognize, is the moment to stop.
Two questions sort out most of what is left. Answer them in order.
Compare the size to the source. Anything that arrived short stopped mid-transfer, and section 11 says why that leaves readers with nothing to work from.
Try to select one word. If nothing highlights, the page is a photograph and section 7 explains what OCR changes about that.
Tracing where this format came from? PostScript files are its direct ancestor, and the EPS format came out of the same printing language.
Working with design files that export to PDF? Illustrator AI artwork writes a PDF-compatible file by default, and CorelDRAW CDR drawings export one on request.
New to all of this? What a file format is covers the ground underneath, and our file format guides list the rest.
Found something here that does not match your version? Send us the exact wording on screen and we will check it against the standard.