Plan a pilot box
Menu

Dublin Core, MARC or EAD: choosing a metadata standard

Reader question: Which metadata standard should our digitized items be described in, and what does our collection software actually need?

By Dastok Editorial TeamIssue: Published

Choose the standard by the system your records will live in, not the other way round. Item records for digitized photographs, letters and objects in Omeka, CONTENTdm or a similar repository are described in Dublin Core. Records in a library catalog are MARC 21 (or MODS in some digital library systems). Finding aids for archival collections are encoded in EAD, with their content written to DACS. Many institutions end up with two of these working together: a finding aid that describes the collection and links to item records for the pieces that were digitized.

Whatever the standard, the mistakes concentrate in the same dozen fields. Dates, creators, rights and identifiers cause most of the trouble when records move between systems, and a crosswalk of those fields, near the end of this article, is worth more than any amount of debate about which standard is best.

This guide is written for small museums, historical societies and libraries that have digitized, or are about to digitize, and need to know what to put in the records. How to get the records into a particular product (file layouts, matching images to records, test loads) is covered separately in importing digitized records into PastPerfect, CONTENTdm or Omeka. How much to describe at item level at all is covered in item-level or collection-level description.

Start from the system your records will live in

Your collection software has already chosen a standard for you, at least in part. Find out what it stores and exports before you design a single field, because the system’s import template is the real target your metadata has to hit.

System What it stores Standard behind it What you should deliver
PastPerfect Four catalogs (objects, photos, archives, library), each with its own fields Its own fields, “based on common standards and practices”; Nomenclature 4 as an optional lexicon for object names Data in PastPerfect’s fields and import layout
CONTENTdm Collections whose fields each carry a Dublin Core mapping Dublin Core, through each field’s “DC map” A spreadsheet in the collection’s field order, with each field mapped to a Dublin Core element
Omeka Classic Items with Dublin Core fields plus item type fields Dublin Core; a plugin adds the full DCMI terms Dublin Core values, plus item type fields where used
Omeka S Items described with linked open vocabularies Dublin Core Terms and Dublin Core Type are built in; others can be imported dcterms values, with controlled vocabularies where configured
ArchivesSpace Collection descriptions (resources) and their components EAD for finding aids; exports also in Dublin Core, MARC and MODS Description written to DACS; finding aids exported as EAD
Library catalog (integrated library system) Bibliographic records MARC 21 MARC records, usually as MARCXML or binary MARC files

A few details from the vendors’ own documentation explain the table. PastPerfect describes its four catalogs as having “fields specific to each type of collection based on common standards and practices” on its product page, and sells an upgrade to Nomenclature 4.0, the lexicon for naming museum objects. OCLC’s CONTENTdm field properties page says each field’s Dublin Core mapping “determines which field in the Dublin Core element set that this field maps to,” and that fields mapped to the same element “can be queried in searches across collections.” (OCLC’s help site calls this the classic version of CONTENTdm, as of November 2026.)

Omeka Classic’s documentation says the Dublin Core fields “are available for every item in your installation,” and its Dublin Core Extended plugin “adds the full set of Dublin Core properties.” Omeka S comes pre-loaded with four vocabularies: Dublin Core, Dublin Core Type, the Bibliographic Ontology and Friend of a Friend. ArchivesSpace’s OAI-PMH documentation lists five metadata formats it can serve: EAD, Dublin Core, DCMI Metadata Terms, MARC and MODS.

The practical rule follows: if the system is built on one standard, describe in that standard and do not fight it. If it holds its own fields, as PastPerfect does, design the fields so they map cleanly to Dublin Core anyway, because sooner or later the records will be shared, harvested or migrated.

What Dublin Core is for

Dublin Core is a small, general set of fifteen elements for describing almost any resource, and it is the common language of digital collections. It suits item-level records for digitized photographs, letters, objects and pages, and it is what most digital repositories and aggregators can read.

The Dublin Core Metadata Element Set takes its name from a 1995 invitational workshop in Dublin, Ohio. Its fifteen elements were first standardized in 1998 as IETF RFC 2413 and later published as ANSI/NISO Z39.85 and ISO 15836. The elements are title, creator, subject, description, publisher, contributor, date, type, format, identifier, source, language, relation, coverage and rights.

DCMI also maintains a larger vocabulary, DCMI Metadata Terms, which repeats the fifteen properties in the dcterms namespace and adds more precise ones such as created, spatial, temporal, extent, medium, isPartOf, rightsHolder, license and accessRights. DCMI says implementers may use either form but “are encouraged to use the semantically more precise dcterms: properties.” If your system offers the fuller terms, use them.

What Dublin Core does well

Dublin Core is easy to learn and easy to share. The element set itself makes no element mandatory, so a record can be as short as a title and an identifier or as full as you can make it. Because so many systems map their fields to it, a Dublin Core record is the most portable description you can create. CONTENTdm’s cross-collection search and the oai_dc format that repositories expose for harvesting both depend on it.

What it leaves to you

Dublin Core says what a field means but not how to fill it. The element definitions are deliberately broad: date is “a point or period of time associated with an event in the lifecycle of the resource,” and creator is “an entity primarily responsible for making the resource.” The standard recommends good practice in comments (an encoding scheme such as the W3CDTF profile of ISO 8601 for dates, a controlled vocabulary for subjects, the DCMI Type Vocabulary for type) but does not require it. Your local rules have to supply the rest: which elements are required, which vocabularies you use, how names are written.

Two more gaps matter for digitized collections. Dublin Core has no built-in way to record where a value came from, such as whether a subject was suggested by a model or entered by a cataloger. And a single record can blur the original object and its digital image, which is why the date and format fields cause so much confusion. Both problems are handled with local rules and extra fields, covered below.

Dublin Core as the sharing format

Even when your main records are richer, Dublin Core is usually the format in which they leave your system. Repositories expose records for harvesting through OAI-PMH, a protocol (version 2.0, 2002) under which repositories “must be able to return records with metadata expressed in the Dublin Core format, without any qualification,” and may offer richer formats beside it. ArchivesSpace, for example, serves Dublin Core and DCMI Metadata Terms alongside EAD, MARC and MODS. CONTENTdm uses its fields’ Dublin Core mappings for searching across collections and offers a standard Dublin Core XML export.

That has a practical consequence. Whatever fields you keep locally, decide how each one maps to Dublin Core, and test what a harvester or a future system will actually see. A local field called “Maker” that maps to nothing will vanish the first time the records are shared.

What MARC 21 and MODS are for

MARC 21 is the record format of library catalogs, and MODS is an XML schema derived from the same tradition for digital library records. Use MARC when the records must live in, or be loaded into, a library catalog; consider MODS when a digital repository asks for richer bibliographic description than Dublin Core.

The Library of Congress describes the MARC formats as “standards for the representation and communication of bibliographic and related information in machine-readable form,” coordinated by the Library with the community. A MARC record is built from numbered fields: 245 for the title statement, 100 for the main entry under a personal name, 264 for production and publication, 300 for physical description, 520 for a summary, 540 for terms governing use and reproduction, 650 for topical subjects, 651 for geographic subjects, 856 for electronic location and access. Each field has indicators and subfields that encode distinctions Dublin Core cannot.

That precision is why MARC belongs in a catalog and is overkill almost everywhere else. A small historical society with no integrated library system has no reason to create MARC records. A library that wants its digitized local history to appear in its catalog beside the printed books does, and its cataloging staff will already know the rules.

MODS, the Metadata Object Description Schema, is “a schema for a bibliographic element set that may be used for a variety of purposes, and particularly for library applications,” maintained by the Library of Congress. Version 3.8 is current. It uses readable element names (titleInfo, name, originInfo, physicalDescription, subject, accessCondition, relatedItem) and sits between Dublin Core and MARC in richness. You will meet it mainly in digital library platforms and in exports, such as the MODS format ArchivesSpace serves.

MARC records for digitized items

When a library catalogs a digitized copy of something it already holds in print, it has two common choices: add the digital version to the existing record, or make a separate record for the reproduction. Either way, two fields carry most of the digital information. Field 856 (electronic location and access) holds the link to the digitized item, and field 533 (reproduction note) describes the reproduction, with its own subfield for the date of reproduction. Ask your cataloging staff which practice your catalog follows before a vendor delivers records, because loading the wrong kind creates duplicates that are tedious to merge.

What EAD and DACS are for

EAD encodes archival finding aids, and DACS says what goes in them. Use them when you describe collections as archives do: as groups of records from one creator, arranged in series and files, rather than as separate items.

Encoded Archival Description is, in the official site’s words, “an XML standard for encoding archival finding aids, maintained by the Technical Subcommittee for Encoded Archival Standards of the Society of American Archivists in partnership with the Library of Congress.” As of the page dated September 10, 2026, the site lists EAD4 as current, still provides EAD3 (version 1.1.1), and says EAD 2002 is “deprecated and no longer supported,” although “still used in the archival community.” EAD4 renames many elements: the title element is unitTitle rather than EAD3’s unittitle, and digital representations are described in a formAvailable element. Ask which version your system imports before anyone writes or converts files.

DACS is the content standard

Describing Archives: A Content Standard is maintained by SAA’s Technical Subcommittee on DACS; the online edition shows version 2022.0.4. DACS does not care about XML. It says what a description must contain, and it marks elements as required, optimum or added value. Required elements include reference code, name and location of the repository, title, date, extent, name of creator (if known), scope and content, conditions governing access, and languages and scripts of the material.

That pairing is the key distinction of this whole subject. DACS is a content standard: it tells you what to write. EAD is an encoding standard: it tells a computer how the writing is structured. Dublin Core and MARC are mostly encoding and element standards, and they need content rules from somewhere else, whether a published standard or your own cataloging manual.

How much of DACS a small institution needs

DACS looks long, but a usable collection description needs only its required elements, a few paragraphs in all. A historical society can write one for each collection in an hour or two from what it already knows: who created or gathered the material, the dates, how much there is, what it covers, and whether anything is restricted. The optimum elements, above all the biographical or administrative history of the creator, add the context researchers value most, and are worth writing for the collections people use. The added-value elements (custodial history, accruals, related materials) can wait until there is time.

Finding aids describe groups, not items

A finding aid describes a collection from the top down: the collection as a whole, then series, then files, and items only where it helps. The Okerlund family papers might be one collection with a correspondence series, a photographs series and the store day book. That structure is the strength of EAD, and it is why archives describe at collection level first. A digitized item can still be reached from the finding aid, through a link at the file or item level.

How the standards fit together

Most institutions with archives and digitized material need two standards at once: a finding aid in EAD for the collection, and item records in Dublin Core for the digitized pieces, each pointing to the other. The finding aid gives context and arrangement; the item record makes each image findable and shareable.

Finding aid (EAD, written to DACS) Item record (Dublin Core) Collection: Okerlund family papers Series 1: Correspondence File: Ida to Marit, 1893 to 1896 Item: letter ACHS-3-012 link to the digital object identifier ACHS-3-012 title Letter, Ida Okerlund to Marit Okerlund creator Okerlund, Ida date 1893-04-12 subject Floods; Bridges machine, not reviewed rights Copyright Not Evaluated isPartOf Okerlund family papers, Series 1 The finding aid gives context and order. The item record makes one image findable.
Fictional example: a collection-level finding aid and an item-level Dublin Core record that point to each other. Field values follow the Dastok metadata mapping worksheet example; element names from DCMI Metadata Terms and the EAD official site, checked November 2026.

The link from finding aid to item is a URL or identifier at the item or file level of the finding aid. In EAD3 it sits in a dao (digital archival object) element; in EAD4 the formAvailable element can carry a URI for “a digital object identified by an ARK (Archival Resource Key) or DOI.” The link back from item to collection is a relation field: dcterms isPartOf in Dublin Core, a relatedItem of type “host” in MODS, field 773 (host item entry) in MARC.

The identifier is what holds this together. Use the same item identifier in the finding aid, the item record and the image file names, and the two descriptions can always be rejoined, even after a migration. Our metadata mapping worksheet asks for the file name stem to be stored in the same record as the accession number for exactly this reason.

When one standard is enough

Not every institution needs both. A museum describing objects one at a time, with no archival collections, may need only its collections management fields and a Dublin Core export for the web. A small historical society can write a short collection-level description following DACS’s required elements in a word processor, publish it as a web page, and keep its digitized items in Omeka with Dublin Core. The structure matters more than the XML.

Spreadsheets are fine, if the columns are standard

Many small institutions catalog digitized items in a spreadsheet before anything goes into software, and that is a sound way to work. Name each column after the element it will become (dcterms:title, dcterms:creator, dcterms:date and so on), keep one row per item, use a single date format, and save as UTF-8 so accented names survive. A spreadsheet built that way imports into Omeka, CONTENTdm and most other systems with little rework. A spreadsheet with columns like “Info” and “Other” has to be re-cataloged.

Content rules and vocabularies

A standard tells you which fields exist; content rules and vocabularies decide what goes in them. Most of the inconsistency in small-institution metadata comes from fields that are filled in different ways by different people, not from choosing the wrong standard.

Write down, for each field, which rule or list applies. The common choices for collections in the United States are:

  • Names: the Library of Congress Name Authority File (LCNAF) for people and organizations that appear there, and a local list in the same inverted form (“Okerlund, Ida”) for those who do not.
  • Topical subjects: Library of Congress Subject Headings (LCSH), or the Thesaurus for Graphic Materials (TGM) for photographs and other visual materials. Both are published by the Library of Congress at id.loc.gov.
  • Object names and materials: Nomenclature 4.0 for museum objects; the Getty Art & Architecture Thesaurus (AAT) for materials, processes and formats.
  • Places: LCSH geographic headings or the Getty Thesaurus of Geographic Names (TGN). Dublin Core’s coverage element recommends TGN by name.
  • Dates: ISO 8601 for exact dates and the Library of Congress’s Extended Date/Time Format (EDTF) for approximate and uncertain ones.
  • Rights: the twelve statements from RightsStatements.org, recorded by their URIs.

Vocabularies are also where machine help fits. Models can suggest subjects and names from recognized text, and our AI cataloging service matches those suggestions to your vocabularies and labels every one until a person reviews it. Dublin Core has no field for that label, so it lives in a separate column or note; the companion article on labeling AI-generated metadata covers how.

The dozen fields where most errors happen

Most metadata problems surface when records move between systems, and they cluster in about twelve fields. The crosswalk below maps each of them across Dublin Core, MARC 21, MODS and EAD3, with the error that most often breaks it.

Field Dublin Core MARC 21 MODS EAD3 Common error
Title title 245 titleInfo/title unittitle Devised titles not marked as devised; titles copied from file names
Creator creator 100, 110 or 111 name with role origination Recipient or subject recorded as creator; names in natural order
Other names contributor (or subject, if the person is a topic) 700, 710; 600, 610 as subjects name with role; subject/name controlaccess (persname, corpname, famname) Photographers, studios and recipients lost or mixed with subjects
Date of the original date (dcterms created) 264 subfield c, plus 008 originInfo/dateCreated unitdate or unitdatestructured “ca. 1895” typed into a date field; digitization date used instead
Date digitized local field or note 533 subfield d (date of reproduction) originInfo/dateCaptured usually omitted at item level Overwrites the date of the original
Topical subject subject 650 subject/topic controlaccess/subject Mixed vocabularies; machine suggestions unlabeled
Place coverage (dcterms spatial) 651 subject/geographic controlaccess/geogname Place of creation confused with place depicted
Type and genre type (DCMI Type Vocabulary) Leader/06, 655 typeOfResource, genre controlaccess/genreform Format of the file entered as the type of the original
Extent and format format (dcterms extent, medium) 300 physicalDescription physdesc or physdescstructured Image file format mixed with the physical medium
Identifier identifier 024 (first indicator 8) or a local field; 856 for the URL identifier; location/url unitid Accession number and file name never stored together
Rights and access rights (dcterms license, accessRights, rightsHolder) 540 (use), 506 (access) accessCondition userestrict, accessrestrict Copyright status mixed with access restrictions; free-text rights
Part of a collection relation (dcterms isPartOf) 773 relatedItem type=“host” the component’s place in the hierarchy; dao for the link No link back to the finding aid or collection record

EAD4 renames several of these (for example unitTitle, unitDate, unitId, accessConditions and useConditions), so check the version your system expects. The metadata mapping worksheet is the place to record your own version of this table, field by field, with the vocabulary and format for each.

A warning about published crosswalks

Crosswalks are useful, but read the notes. The Library of Congress’s MARC to Dublin Core crosswalk, issued April 24, 2008, maps MARC’s 100, 110, 111, 700, 710 and 711 name fields to Dublin Core contributor in its unqualified version, with the note “Creator element not used.” It also warns that “it is not expected that round-trip mapping is possible,” because “once MARC data is converted to Dublin Core, not enough information is retained to allow for mapping back to MARC accurately.” If you convert catalog records to Dublin Core for a repository, check where the creators went, and keep the MARC records as the source.

Dates, names, rights and identifiers in detail

Dates, names and rights account for most of the errors in the table above, because each mixes a value with a judgment about it. Type, format and identifiers follow close behind. Handle each with a written rule and a field for the judgment, not just the value.

Dates

Keep three things apart: the date of the original, the date it was digitized, and any date the item is about. A photograph taken about 1895, scanned in 2026, showing a flood of 1893, has three dates, and a single date field cannot hold them.

For the date of the original, use a machine-readable form plus a display form if your system has one. Dublin Core recommends an encoding scheme “such as the W3CDTF profile of ISO 8601,” which writes April 12, 1893 as 1893-04-12. For uncertain dates, the Extended Date/Time Format adds qualifiers: “1984?” for an uncertain year, “2004-06~” for an approximate year and month, and “X” for unspecified digits, as in “201X.” EDTF’s level 0 covers features of ISO 8601-1; levels 1 and 2 cover features described in ISO 8601-2.

MODS gives the same advice in its own terms. Its originInfo guidelines say that when a date is uncertain or supplied by the cataloger, record that “through the use of the qualifier attribute rather than inserting characters such as ‘ca.’, brackets, or a question mark as part of the date string.” A date field that holds “ca. 1895?” cannot be sorted, filtered or converted. Put “about 1895” in a display field and 1895~ (or a qualifier) in the machine field.

Archives add one more distinction: inclusive dates (the full span of a collection) and bulk dates (the span most of it covers). EAD4’s unitDate has an attribute “to indicate if unitDate represents inclusive dates or bulk (predominant) dates.” The Okerlund papers might be inclusive 1893 to 1896, with the store day book dated more narrowly.

Names

Record names in inverted form from an authority file or a local list, and record the role. The creator of a letter is its writer; the recipient is a contributor or a subject, depending on your rules, but never the creator. The creator of a cabinet card is the studio or photographer; the people in it are subjects. The most common error in small-institution records is the person a collection is named after appearing as the creator of everything in it.

MARC records roles with relator terms or codes, MODS with a role element, and Dublin Core only with the choice between creator and contributor. If roles matter to your users (photographer, correspondent, publisher), keep them in a local field or a note, because simple Dublin Core will flatten them.

Rights

Keep copyright status and access restrictions in separate fields, and use a standard statement for copyright. RightsStatements.org “currently provides 12 different rights statements” for cultural heritage institutions, in three groups: in copyright, no copyright, and copyright status unclear. Each has a URI, such as the statement for “Copyright Not Evaluated,” and systems can display and filter on the URI. Record the URI, not just the label.

Access restrictions are a different thing: a donor’s condition, privacy for living people, a culturally sensitive item. They go in dcterms accessRights, MARC 506, EAD3’s accessrestrict or EAD4’s accessConditions. Conditions on use (publication, reproduction) go in MARC 540 or EAD’s userestrict or useConditions. Mixing the two in one free-text field is how restricted material ends up published.

Type and format

Type describes what the original is; format describes its physical or digital form. Dublin Core recommends the DCMI Type Vocabulary for type, a short list of classes such as Text, Image, Sound, Moving Image and Physical Object, and the list of Internet media types for format. Omeka builds the Dublin Core Type classes into every installation.

The trouble comes from describing the original and the digital image in one record. A letter is Text, even though what the website delivers is a JPEG. A glass plate negative is an Image whose physical medium is glass, and whose access file happens to be a TIFF or JPEG. Decide whether your item records describe the original (most collections do, with the digital file as a property of the record) and apply it everywhere. Record the physical medium in format or medium, and leave file formats to the technical metadata your system or your preservation package keeps.

Identifiers

Every item needs one identifier that never changes and is never reused, and that identifier should appear in the item record, the finding aid and the file names. Accession numbers usually work, because the institution already controls them. Shelf locations and box-and-folder positions do not, because they change when collections are rehoused.

Where you publish online, add a stable web address for each item and keep it working through system changes. EAD4’s guidance mentions persistent schemes such as the ARK and the DOI for linking to digital objects; a simple, permanent URL pattern on your own domain serves the same purpose for most small institutions, as long as someone is responsible for redirects when the system changes.

Choosing for your situation

The right combination depends on what you hold and where you publish. Four common situations cover most small institutions.

  1. Historical society with mixed collections, publishing online. Describe each collection at collection level following DACS’s required elements, even if only as a web page. Describe digitized items in Dublin Core in Omeka or CONTENTdm, with dcterms where available. Link the two by identifier.
  2. Museum with objects and some archives, on collections management software. Keep objects in the system’s catalog fields with Nomenclature for object names, and map those fields to Dublin Core for the web and for harvesting. Describe archival collections at collection level, linking item records where digitized.
  3. Public or academic library with special collections. Keep printed and published material in MARC in the catalog. Describe manuscript collections in EAD to DACS, in ArchivesSpace or a finding aid site. Put digitized items in the digital repository in Dublin Core or MODS, linked to both. Our page on libraries and special collections covers how digitization usually fits that setup.
  4. County or municipal records office. Describe record series at collection level, with access restrictions recorded carefully, and describe digitized items only where public use justifies it. Retention and legal requirements come first; the metadata standard follows them.

If you plan to contribute records to a regional or national aggregator, ask for its written metadata requirements before you design your fields: which elements it requires, which date formats it accepts and which rights statements it displays. It is far easier to meet those requirements at the start than to retrofit thousands of records.

In every case, decide the twelve fields above before the first scan. Metadata decided after capture tends to be whatever the file names happened to say.

Example (fictional). The Aster County Historical Society publishes its digitized items in Omeka Classic and has no archival software. For Box 3 it writes a one-page collection description for the Okerlund family papers following DACS’s required elements: reference code, repository, title, dates (1893 to 1896), extent (one box, 69 items), creator (the Okerlund family), scope and content, conditions governing access and languages. It publishes the page on its website.

Each digitized item gets a Dublin Core record. Letter ACHS-3-012 has the identifier ACHS-3-012 and the devised title “Letter, Ida Okerlund to Marit Okerlund.” Its creator is “Okerlund, Ida,” the date 1893-04-12, and the subjects “Floods; Bridges,” flagged as machine suggestions until a cataloger reviews them. Rights carry “Copyright Not Evaluated” until the rights review is done, and isPartOf names the collection and series. A cabinet card dated from its mount style gets 1895~ in the date field and “about 1895” in the description.

The collection page links to an Omeka search for the box, and every item record links back to the collection page. When the society later moves to a system with EAD support, the collection description and the identifiers carry over unchanged.

The fictional example collection shows these records alongside the images and text they describe.

What to do next

You can settle most of this in an afternoon with the people who catalog and the person who runs your collection system. A practical sequence:

  • Find out what your system imports and exports, and in which standard and version (Dublin Core or dcterms, MARCXML, MODS, EAD3 or EAD4).
  • Decide whether you need collection-level description, item-level description or both, and how they will link.
  • Fill in a field map for the twelve fields above: your field name, the target element, whether it is required, and the vocabulary or format. The metadata mapping worksheet has the columns for this.
  • Write the date, name and rights rules down in one page, with examples, before anyone starts entering data.
  • Add a field or column for where each value came from (person or machine) and whether it has been reviewed.
  • Test the map with ten real records exported from the system and imported somewhere else, then fix the fields that broke.

If you are planning a digitization project and want the records built to your system’s template from the start, our AI cataloging service delivers Dublin Core, MARC or EAD with every machine-suggested field labeled. A pilot box shows the records for a small batch of your own material before anything larger is decided.

Working through this with a real collection?

If this article describes a decision you are facing, tell us about the collection. A pilot box answers most questions with your own material.

Plan a pilot box

Studio and contacts

24 South Weber Street, Suite 225
Colorado Springs, CO 80903
(719) 244-8450sales@dastok.com