PDF/A-1b vs PDF/A-2b vs PDF/A-3b: Which One Should You Use?

For most new documents, PDF/A-2b is the sensible default: it is based on PDF 1.7, allows JPEG 2000 and transparency, and is accepted by the FDA and EMA for submissions. Choose PDF/A-1b only when a recipient requires it, and PDF/A-3b only when you must embed non-PDF files, such as an XML invoice, inside the archive copy.

All three are parts of ISO 19005, the PDF/A family of standards for long-term preservation. The number tells you which part of the standard, and therefore which version of PDF it builds on. The letter tells you the conformance level. Getting both right matters when a regulator, an archive or a customer specifies one exactly.

The three parts at a glance

ISO's catalogue lists all three as current editions. A fourth part, PDF/A-4 (ISO 19005-4:2020), is based on PDF 2.0 and is covered briefly below.

What the "b" means

The letter is the conformance level. The Library of Congress format descriptions summarise them like this:

• Level b (basic) is minimal conformance to ensure that the rendered visual appearance of the file can be preserved over the long term. It does not require the document's logical structure.
• Level a (accessible) meets every requirement in the standard, including those for structural and semantic properties, so the reading order and structure are recorded in the file.
• Level u (Unicode), introduced with PDF/A-2, is level b plus the requirement that all text has Unicode equivalents, which makes it reliably searchable and extractable.

Level b is the one used most for scanned paper. The Library of Congress notes that a PDF/A-1b file may contain only scanned images with no text at all, and that OCR text supporting search is permitted only if it is invisible.

What every PDF/A part has in common

All PDF/A files are meant to be self-contained, so they render the same way in decades without anything outside the file. Across PDF/A-1 to PDF/A-3, the Library of Congress lists these core rules:

• Encryption is not allowed, so a password-protected file cannot be PDF/A.
• All fonts must be embedded, and must be legally embeddable.
• JavaScript and launching executable files are prohibited.
• Audio and video content are forbidden.
• There must be no external dependencies; required metadata schemas must be embedded too.

PDF/A-1b: the oldest and most restrictive

PDF/A-1 was published in 2005 and is a constrained form of PDF 1.4. It does not allow several features that PDF/A-2 later added, including JPEG 2000 compression, compressed object streams and transparency in graphics, so scanned documents tend to be larger.

It is still very much alive. PDF/A-2 was not intended to replace it, and the Library of Congress quotes the PDF/A Association's statement that PDF/A-1 remains in effect without limitation. Some archives prefer it: the Library of Congress reports that the British Library's PDF/A assessment recommends preferring PDF/A-1 to reduce the risk from attached files.

Choose PDF/A-1b when a recipient, records system or contract asks for it by name, or when you need the widest compatibility with older archiving systems.

PDF/A-2b: the practical default

PDF/A-2 moved the base to PDF 1.7. The Library of Congress lists what this added:

• Improvements to tagged PDF for accessibility.
• Compressed object and cross-reference streams, for smaller files.
• Embedding of PDF/A files as attachments, plus portable collections.
• Transparency for graphical elements.
• JPEG 2000 compression.
• Digital signatures in line with the PAdES standard.

JPEG 2000 support is the reason PDF/A-2b is often recommended for scanning paper: the Library of Congress notes this has led to recommendations for PDF/A-2b because of the reduction in file size.

Choose PDF/A-2b when you have no specific instruction. It suits office documents converted from Word or Excel, scanned records, and anything that may be signed. If the text needs to be reliably searchable, consider PDF/A-2u; if the document must be accessible, PDF/A-2a.

PDF/A-3b: PDF/A-2 plus any embedded file

ISO 19005-3 describes PDF/A-3 as PDF 1.7 for preserving the static visual representation of documents, in addition to allowing any type of other content to be included as an embedded file or attachment. The Library of Congress puts it more simply: PDF/A-3 is identical to PDF/A-2 except that it allows files in any format to be embedded.

Each embedded file has to be explicitly associated with the document or part of it, with a declared relationship such as Source, Data or Alternative, and a MIME type. One important motivation was electronic invoicing: the German ZUGFeRD and French Factur-X formats carry a machine-readable XML invoice inside a human-readable PDF/A-3.

There is a catch for preservation. The Library of Congress notes that embedding arbitrary files poses challenges for archives, both when files are received and over the long term. It also reports that several national archives list PDF/A-1 and PDF/A-2 as preferred formats but do not list PDF/A-3, and that a 2014 National Digital Stewardship Alliance report recommended PDF/A-2 rather than PDF/A-3 for documents with no embedded files, or with only PDF/A files embedded.

Choose PDF/A-3b only when the embedded file is part of the requirement, as with structured e-invoices, and you have a separate plan for the embedded format.

What regulators accept

For medicines submissions the answer is clear. The FDA's PDF Specifications (version 4.1) say PDF versions 1.4 through 1.7, PDF/A-1 and PDF/A-2 are acceptable. The EMA's harmonised eCTD guidance (version 6.0.1, section 2.9.2) lists the same set, and section 2.9.3 adds that PDF/A format is recommended where possible. The EU's accepted file formats for eCTD v4.0 list PDF/A-1, PDF/A-2 and PDF 1.4 to 1.7 as well.

PDF/A-3 is not on any of those lists, and the FDA specification also says submitted PDFs must not contain attachments. For an eCTD, choose PDF/A-1b or PDF/A-2b. Our guide on preparing PDFs for regulatory submissions covers the other requirements.

A note on PDF/A-4

ISO 19005-4:2020 bases PDF/A on PDF 2.0 and drops the a, b and u levels. It has two named profiles: PDF/A-4f, which allows embedded files of any type and succeeds PDF/A-3, and PDF/A-4e for engineering documents. The ISO catalogue shows the 2020 edition is due to be replaced by a revision under development. Check that your recipients and validation tools support it before adopting it.

How to check which PDF/A a file really is

A PDF can declare itself PDF/A in its metadata whether or not it actually conforms. The only reliable answer comes from a validator. veraPDF is a free, open-source validator that covers all PDF/A and PDF/UA parts and conformance levels. Run it on the final file, not on a draft, because later changes can break conformance.

A quick clue is the PDF version, which View Metadata shows. PDF/A-1 is based on PDF 1.4 and PDF/A-2 and PDF/A-3 on PDF 1.7, so the version can hint at which part a file was made to, but it tells you nothing about conformance, and View Metadata does not read the PDF/A identification, so treat it only as a first look.

Be careful with edits made after conversion, because they can break conformance. For example, our OCR PDF tool adds its invisible text layer in the standard Helvetica font without embedding it, and PDF/A requires every font to be embedded. Run OCR before conversion, then validate the converted file.

For records in regulated work

If your quality system archives SOPs, batch records or reports as PDF/A, write the exact part and level into the procedure, say which validator is used and what happens when a file fails. A file that merely "looks like" PDF/A in a viewer has not been checked. Keep the source files under document control too, because PDF/A preserves appearance, not editability.

This article is general information about the PDF/A standards, not regulatory or legal advice. Confirm requirements with the receiving agency, archive or your QA team.