Describe what you can see and what the object itself tells you, and record how you know each thing. Give the photograph a devised title in brackets, write uncertain dates in one consistent form such as “approximately 1895” or the machine form 1895~, name people only when an inscription or a documented source supports it, and say “unidentified” when nothing does. Record the process and mount with controlled terms, keep every earlier attribution with its source instead of overwriting it, and treat machine suggestions about identity, place or date as leads for a person to check, never as facts.
Most historical societies hold boxes of portraits with nothing on the back. This article sets out a method for those photographs: the fields to fill, the wording to use when you do not know, how to date a print from its process and mount, how to record identifications that arrive later from descendants and researchers, and where software helps and where it must not decide. It uses one fictional collection, the Aster County Historical Society’s Box 3, throughout.
Start from evidence: what is known and how you know it
A good record of an unidentified photograph separates three things: what is visible in the image, what the object says about itself, and what someone has concluded. Each statement in the record should be traceable to one of those.
Describing Archives: A Content Standard (DACS), the Society of American Archivists’ content standard for archival description, says to take date information “from any reliable source, including the internal evidence of the materials being described.” The same idea applies to names and places. The evidence on a photograph usually falls into a short list:
| Evidence | Where to look | What it can support |
|---|---|---|
| Inscriptions | Back of the print or mount, album page, sleeve, envelope | Names, places, dates, occasions |
| Studio imprint | Front or back of the mount, embossed or printed | Photographer, town, a date range from the studio’s years |
| Mount style | Card size, color, edges, corners, borders | A date range for card photographs |
| Process | The image itself under magnification and raking light | A date range, and the right format term |
| Content | Clothing, vehicles, signs, buildings, landscape | A date range and sometimes a place |
| Provenance | Accession records, donor letters, the folder it came in | Family, place and period of the group |
| Related records | Letters, diaries, directories, newspapers in the same collection | Names and dates that match the image |
The North Dakota State Library’s guide to dating photographs sorts the same clues into four groups: format, details in the image, photographer or studio, and the people themselves. It also makes the point that matters most for cataloging: “Even a date range or an approximation (like ‘between 1925 and 1935’ or ‘late 1940s’) is better than nothing.”
Record only what you have, and say where it came from
The University of North Texas Libraries’ instructions for photograph records set a useful default: “only include information that you have available (vs. doing research, adding contextual notes, or making assumptions).” Research can come later and be added with its source. A record that mixes fact and guesswork without saying which is which is harder to correct than one that leaves a field empty.
In practice that means a short note behind every uncertain statement: “date from mount style,” “name from pencil inscription on back,” “place from studio imprint.” The note takes seconds to write and saves the next cataloger from repeating the work or trusting a guess.
The fields to fill for every photograph
A record for a historical photograph needs the same core fields whether or not anyone is identified. The table lists them, with what to write when the answer is unknown.
| Field | What to record | When it is unknown |
|---|---|---|
| Identifier | The item number, matching the file name of the scan | Never unknown; assign one at accession |
| Title | A devised title in square brackets if the photographer gave none | “[Portrait of an unidentified man]” |
| Photographer or studio | Name from the imprint or a documented source | Leave empty; do not guess from style |
| Date | A single date, a range or an estimate, in one consistent form | “undated,” or an estimate with its basis |
| Date basis | How the date was reached | “No evidence” if none |
| People shown | Names, with how each was identified | “Unidentified” plus visible features |
| Place | Where it was taken, with the basis | Leave empty, or “studio portrait” if that is all you know |
| Format and process | Controlled terms, such as cabinet photographs and albumen prints | The broader term, such as photographic prints |
| Physical description | Image and mount dimensions, color, condition | Never unknown; measure it |
| Inscriptions | Exact transcription, with location and medium | “No inscriptions” |
| Description | What is visible, beginning “Photograph of” or “Portrait of” | Describe what you see |
| Subjects | Controlled subject terms for what the image shows | Choose broader terms |
| Provenance | Donor, accession, original folder or album | “Source unknown” |
| Identification history | Each identification, who made it, when and on what evidence | Empty until someone offers one |
The last field is the one most systems lack. If your catalog has no field for it, use a note field with a consistent label, so it can be found and moved later.
Link photographs to the documents around them
Photographs rarely arrive alone. A letter may mention a portrait sitting, a ledger may record a payment to a photographer, and a clipping may describe the event in a group photograph. When a document in the same collection bears on a photograph, link the two records and say what the document shows. The link is evidence in its own right, and it survives even if the photograph’s identification later changes.
Set a time limit for research on each photograph. Most unidentified portraits will not be solved in one sitting, and a record with an honest “unidentified” and good physical description is complete enough to publish. The identification can come later, from the people who see it online.
Physical description is never unknown
Measure the image and the mount separately, note the color and edges of the mount, and record the process if you can identify it. These facts are certain, they help with dating, and they let a researcher match the print to one in another collection. A cataloger who cannot name a single sitter can still produce a record that is accurate and useful.
Titles and descriptions for people you cannot name
When the photographer gave no title, write a short devised title in square brackets that describes what is shown. Use “unidentified” for people whose names you do not know, and add a visible detail that tells this photograph apart from the others.
UNT’s guidance is specific. If the photograph has no official title, “create a title that briefly describes the photograph” and “enclose the title in [square brackets] to show that it was assigned by the editor and not the photo’s creator.” For portraits of unknown people, it suggests titles that “include age ranges or noticeable features like a uniform.” Its descriptions start with “Photograph of,” and should “focus on what is in the image rather than context or assumptions about what might be happening.”
Wording that stays accurate
| Instead of | Write | Why |
|---|---|---|
| [Grandma Okerlund?] | [Portrait of an unidentified older woman] | A guess in a title spreads into every search result |
| [Unknown family] | [Group portrait of a man, a woman and three children] | “Family” is a conclusion; the image shows people together |
| [Wedding photo] | [Portrait of a young man and woman, standing] | Unless the dress, flowers or an inscription show a wedding |
| [Miner] | [Portrait of an unidentified man in work clothes holding a lamp] | Describe the object, not the occupation you infer |
| [Indian woman] | [Portrait of an unidentified woman in a beaded dress] | Do not assign identity or community from appearance |
The last row matters beyond accuracy. Identity, community and relationships inferred from appearance are often wrong and can cause harm. If a community or family can identify the person, record what they tell you with its source, as described below.
Description: what is visible, then what is written
Write one to three sentences on what the image shows: the number of people, their positions, clothing and props, the setting or painted backdrop, and anything legible. Then transcribe every inscription exactly, in its own field or note, with its location and medium, as in this note from the fictional Box 3: “Back of mount, pencil: Ida and Marit.” Do not correct spelling in a transcription; note “[sic]” sparingly if a reader might think the error is yours.
Writing uncertain dates the same way every time
Choose one form for estimated dates and use it everywhere. DACS gives a human-readable set of forms, and the Extended Date/Time Format (EDTF) gives a machine-readable one; many catalogs store EDTF and display words.
DACS rule 2.4.15 says that if no date can be found or determined, “estimate the nearest year, decade, century, or other interval as precisely as possible. Record estimated dates in a consistent fashion.” Its examples include “probably 1867,” “approximately 1925,” “before 1867,” “after 1867,” “1892 or 1893” and “1890s.” When no estimate is sensible, rule 2.4.16 says to use “undated” and not the abbreviations “n.d.” or “s.d.” A footnote recommends spelling out words such as “approximately” rather than abbreviating them, because “abbreviations may not be understood by all users.”
The EDTF specification, published by the Library of Congress on February 4, 2019, defines short symbols for the same ideas. The characters “?”, “~” and “%” mean “uncertain,” “approximate” and both. An “X” replaces unknown digits, so 189X is a year in the 1890s. Square brackets hold “one of a set,” so [1892,1893] means one of those two years, and two dots mark an open end.
| What you know | DACS wording | EDTF form |
|---|---|---|
| The exact day, from an inscription | 1896 March 14 | 1896-03-14 |
| About a year, from the mount style | approximately 1895 | 1895~ |
| Probably a year, but not certain | probably 1895 | 1895? |
| Some year in a decade | 1890s | 189X |
| One of two years | 1892 or 1893 | [1892,1893] |
| A span of years, both ends estimated | approximately 1890-1899 | 1890~/1899~ |
| No later than a year | before 1900 | [..1899] |
| No evidence at all | undated | (leave empty) |
UNT’s photograph instructions use the same EDTF notation in practice: a tilde at the end for approximate dates, “1940~/1948~” for an approximate range, an “X” for a known decade such as 197X, and the “one of a set” brackets for a photograph taken on one of several dates. Its system shows a warning if an editor enters a date that does not conform to EDTF.
Record the basis of every estimate
A date without a basis cannot be checked or improved. Add a short note: “Date estimated from cabinet card mount and albumen print,” or “Date from the school term in the inscription.” When better evidence arrives, the next cataloger can see what the old estimate rested on and replace it with confidence.
Folders, albums and copy prints
The same forms work above the item level. DACS asks for inclusive dates for a group, the earliest and latest, and says that where those dates are estimates, to “indicate the estimated dates in a clear and consistent fashion,” with examples such as “circa 1870-1879.” A folder of unidentified portraits can carry an estimated range even when no single print can be dated closely.
Copies need their own care. UNT’s instructions note that if a photograph “is a copy photo or reproduction, the creation date should be when that item was made, not the original.” DACS similarly says to record the details of a reproduction, including its date, in the scope and content note. A 1950s copy print of an 1890s portrait therefore gets a 1950s creation date, with the date of the image it reproduces described separately.
Dating from process, mount and studio
Process and format give the firmest date limits for an unidentified photograph: the earliest year a print could have been made, and a period when it was common. Narrow the range further with the studio’s years in business and the details in the image.
The Library of Congress Thesaurus for Graphic Materials (TGM) includes short scope notes with date ranges for common formats. As served by the Library’s vocabulary service and checked in November 2026, they read:
- Cartes de visite: card photographs with mounts of about 4 x 2.5 inches, “Introduced in the United States in 1859; made into the 1900s.”
- Cabinet photographs: card photographs with mounts of about 6.5 x 4.25 inches, “Introduced in the United States in 1866; popular until ca. 1900.”
- Albumen prints: the “Predominant paper print photographic process in the 1800s; popular 1850s-1890s.”
- Tintypes: “Popular mid-1850s through 1860s; in use through 1930s.”
- Gelatin silver prints: “introduced in the 1880s; dominant black-and-white photographic print process since the 1890s.”
Sources disagree, so cite the one you used
Authoritative references do not always agree on these dates. The Getty Art and Architecture Thesaurus says cabinet photographs “were introduced by F. R. Window in 1867” and that the size “remained popular until World War I.” TGM gives 1866 for the United States and “until ca. 1900.” The difference is small, but it shows why a date note should name its source: “cabinet card, so probably 1866 or later (TGM).”
Combine the clues
Each clue gives a range; the date is where the ranges overlap. An albumen print on a cabinet mount sits inside both ranges, so roughly 1866 to the 1890s. A studio imprint narrows it further if city directories or newspaper advertisements show when that studio worked at that address. Clothing, props and backdrops narrow it again, with less certainty, so record them as supporting evidence rather than as the basis of the date.
Examining a print, step by step
- Handle by the edges, over a clean table, with the print supported on a sheet of board.
- Read the back first. Inscriptions, stamps, labels and the studio’s printed back are often the best evidence on the object.
- Measure the image and the mount separately, and note mount color, edges and any printed border.
- Look at the image surface under magnification and low, raking light to help identify the process; record “photographic print” if you are not sure.
- Note the setting: painted backdrop, posing chair, balustrade, props. These help group prints from one studio.
- Describe clothing and objects briefly, as supporting evidence for a date, not the basis of it.
- Write the date and its basis before moving on, while the evidence is in front of you.
Build your own reference set
Your own dated photographs are the best guide to your undated ones. Every print in the collection that carries a dated inscription, or comes from a dated album or letter, is a reference point for its studio’s mounts and backdrops. Photograph or list those reference prints by studio, with the date and its source, and keep the sheet beside the cataloging station.
Over a few boxes, the sheet starts to answer questions on its own. A backdrop that appears on a card inscribed 1894 and another inscribed 1896 is good evidence that an unmarked card with the same backdrop came from that studio in that period. Record such conclusions as supporting evidence with the reference prints named, so others can check the comparison.
Researching the studio
A studio imprint opens a line of research that often narrows the date more than any visual clue. City and business directories, newspaper advertisements and published directories of regional photographers can show when a photographer worked and at which address. The North Dakota State Library’s dating guide, for example, links to its own index of North Dakota photographers and to directories for other regions of the United States and Canada. Record the source and the years found, as in this fictional note: “Halvorsen Studio listed at Main Street, Wren Gulch, in the 1892 and 1897 directories.”
Use the widest honest range rather than a precise guess. “Approximately 1866-1899, from albumen print on cabinet mount” is more useful than “1885” with no basis, because a researcher can trust the first and test the second.
Names and places: identified, attributed or unidentified
Record each person in a photograph at one of three levels: identified (documented by a reliable source), attributed (a reasoned suggestion with named evidence), or unidentified. Say which level applies, and who made the identification, when and on what basis.
| Level | Meaning | Example of the evidence | How to record it |
|---|---|---|---|
| Identified | Documented by a source you would cite in print | An inscription naming the sitter, with its hand and date noted if known; a captioned copy in a dated publication | Name in the people field, source in a note |
| Attributed | A reasoned identification that is not documented | A descendant’s recognition; comparison with an identified portrait | “Attributed to” or “possibly,” with who, when and why |
| Unidentified | No supporting evidence | None | “Unidentified” plus visible features |
Keep earlier attributions in the record
When a new identification replaces an old one, do not overwrite the old one. Move it to the identification history with its source and date, and add the new one beside it. Researchers who saw the old name in a published book or an earlier finding aid can then follow the change, and if the new identification is later doubted, the evidence for both is in front of the next cataloger.
A simple history entry carries five parts: the name proposed, who proposed it, when, the evidence, and the decision. A fictional example: “Proposed as Ida Okerlund by a descendant, 2026-05-12, from a family album copy labeled ‘Ida 1895’; accepted as attributed, 2026-05-20, by the collections manager.”
Group photographs: identify by position
In group portraits, identify people by position, using a fixed order such as rows from front to back and people from left to right, as the viewer sees them. UNT’s guidance asks whether names are known and suggests noting them “pictured from left to right.” Record each position separately, so one person can be identified while the others stay unidentified.
Inscriptions on group photographs often name some people without saying which is which. The back of a school photograph may name the teacher; the adult standing beside the pupils is probably that teacher, but the inscription does not say so. Record the name from the inscription in the description, and record the position-level identification as attributed, with the reasoning written out.
Places follow the same rule
A studio imprint tells you where the photographer worked, not necessarily where the sitter lived. A painted backdrop says nothing about place at all. Record a place only with its basis, such as “Wren Gulch (Colo.), from studio imprint” in the fictional example collection, and for outdoor views, describe the landmarks that support the identification.
Controlled terms for formats, processes and subjects
Use a published vocabulary for format and process so the same object is described the same way across your catalog and others. For photographs, the two vocabularies most often used in the United States are the Library of Congress TGM and the Getty AAT.
The Library of Congress describes TGM as “a tool for indexing visual materials by subject and genre/format,” and its research guide explains how to apply the genre and format terms. The AAT, from the Getty Research Institute, offers persistent identifiers for each concept. Recording the identifier as well as the term means a later system can link the record to the vocabulary without guessing.
| Term | TGM identifier | AAT identifier |
|---|---|---|
| Cabinet photographs | tgm001517 | 300127131 |
| Cartes de visite | tgm001729 | |
| Albumen prints | tgm000227 | 300127121 |
| Tintypes | tgm010835 | |
| Gelatin silver prints | tgm004499 | 300128695 |
| Group portraits | tgm004759 |
Identifiers were checked against each vocabulary’s data service in November 2026; blank cells were not checked for this article. For rare or high-value prints, the RBMS standard Descriptive Cataloging of Rare Materials (Graphics), the successor to Elisabeth Betz Parker’s 1982 Graphic Materials rules, gives fuller rules for transcription and physical description.
Subjects: describe the image, not the guess
Subject terms should describe what the image shows: “Portraits,” “Group portraits,” “Schools,” “Bridges.” A subject such as “Okerlund family” belongs only on records where the Okerlunds are identified or attributed, and the attribution note should say so. For how detailed each record needs to be, and when a folder-level record is enough, see item-level or collection-level description.
Asking the community, and recording what they tell you
Descendants, local historians and long-time residents can often name people and places no cataloger could. Invite them to help, give them a way to say how they know, and record each suggestion as an attribution until it is checked.
A simple identification form
Whether you use an online form, a comment field or a sheet at a reunion table, ask for the same things:
- Which photograph, by its identifier.
- Who or what they recognize, and where in the image (left to right, front row).
- How they know: family knowledge, a matching photograph, an inscription elsewhere, a published source.
- Any copy or document they can share or describe.
- Their name and contact, and whether they agree to be credited.
Keep every suggestion in the identification history, including the ones you do not accept. A rejected name, with the reason, stops the same suggestion from being accepted later by someone who does not know it was already checked.
| Item | Date | Name proposed | Proposed by | Evidence | Decision |
|---|---|---|---|---|---|
| ACHS-3-C09 | 2026-03-10 | Okerlund, Walter | Face-match software | Similarity score only | Rejected: no documented portrait to compare |
| ACHS-3-C11 | 2026-06-02 | Lund, Thea (front row, right) | Volunteer, from inscription | “Miss Thea Lund, teacher” on back; only adult present | Attributed |
| ACHS-3-C11 | 2026-07-19 | Lund, Thea | Descendant | Family copy of the same print, labeled “Aunt Thea at school” | Raised to identified |
The rows above are fictional. Then review each suggestion. A match with an identified portrait elsewhere, or a family album copy with a label, can raise a suggestion to “attributed.” A single memory may stay “possibly” until something supports it. Credit contributors in the record when they agree to it; it encourages the next contribution.
Privacy and sensitive images
Not every photograph should be put online for identification. The SAA Core Values Statement and Code of Ethics says archivists “establish procedures and policies to protect the privacy interests of the donors, individuals, groups, and organizations whose public and private lives and activities are documented in archival holdings.” It adds that they restrict access where appropriate to protect privacy, “particularly for individuals and groups who have had no voice or role in collections’ creation, retention, or public use.”
Take particular care with photographs of people who may still be living, of children, of medical or institutional settings, and of communities whose own protocols govern their images. For material from Native American communities, our article on digitizing Native American cultural materials covers consultation and community control.
Where AI helps and where it must not decide
Software can do useful first passes on a photograph collection: read inscriptions, suggest subject terms, propose a date range from visual cues, and group images that may show the same person or the same studio backdrop. It must not decide who a person is or where a photograph was taken. Those conclusions need a person, a named source and a record of the evidence.
Useful jobs for machines
| Task | What the machine produces | What a person does |
|---|---|---|
| Reading inscriptions | Text from the back of mounts, by handwriting recognition | Checks the text against the image and corrects names |
| Suggesting subjects | Draft terms such as portraits, schools, bridges | Accepts, changes or rejects each term |
| Grouping | Sets of images that look alike: the same sitter, backdrop or mount | Decides whether each group is real and what it means |
| Dating cues | A proposed range from clothing, process or mount | Treats it as a lead and records the basis if accepted |
| Finding candidates | A ranked list of similar faces from identified portraits | Examines the evidence and decides; records the method |
Grouping is the most valuable of these for unidentified portraits, because it turns one mystery into a set. If a known sitter appears in one image of a group, the others become candidates for comparison, never automatic matches. A team at Virginia Tech built a public site, Civil War Photo Sleuth, around this pairing of software and people; its researchers’ peer-reviewed papers have titles such as “Photo Sleuth: Combining Human Expertise and Face Recognition to Identify Historical Portraits” (2019) and “DoubleCheck: Designing Community-Based Assessability for Historical Person Identification” (2023).
Why identity is different
Face recognition returns similarity, not identity. The National Institute of Standards and Technology (NIST), in its 2019 study of demographic effects, tested 189 algorithms from 99 developers and found “empirical evidence for the existence of demographic differentials in the majority” of them. It explained the stakes of a false positive in a one-to-many search: it “puts an incorrect match on a list of candidates that warrant further scrutiny.”
NIST’s test photographs came from operational government databases, “18.27 million images of 8.49 million people,” not from faded nineteenth-century prints. Its results therefore say nothing direct about how a given tool performs on a cabinet card from 1895, which is one more reason to measure any tool on your own identified portraits before trusting its suggestions, and to keep every suggestion labeled as machine output.
Test a tool on photographs you can already name
Before using any face-matching or dating tool on unidentified portraits, run it on portraits whose sitters and dates are documented. That gives you a measured result on your own material instead of a vendor’s description.
- Pick a test set of identified, dated portraits from the same period and formats as the unidentified ones, including several people who appear more than once.
- Hide the answers from the tool and from the person scoring the results.
- Run the tool and keep every suggestion it makes, with its score.
- Count how often the right person appears at the top, how often in the first few suggestions, and how often a wrong person ranks above the right one.
- Decide how you will use it. If wrong matches rank high often, the tool can still help group images, but it should not be used to suggest names.
Write down the test set, the date and the result, and repeat the test when the tool or its model changes.
Labeling and review
Label every machine-suggested field as machine output until a person has reviewed it, and record which fields were reviewed and by whom. Our article on labeling AI-generated metadata gives field-level patterns for doing this, and our AI policy explains how Dastok labels machine output in the records it delivers. For identity, add one more rule: a face-match suggestion never moves a person from “unidentified” to “identified” by itself. At most it supports an attribution, with the method named, after a person has examined other evidence.
This article is not legal advice. Before running face recognition on photographs of people who may still be living, ask counsel whether any privacy law applies to that use in your state.
A worked record and a checklist for the next box
The example below applies the method to two cabinet cards from the fictional Box 3: one with an inscription and one with nothing on it.
Example (fictional). Two cabinet cards from the Aster County Historical Society’s Box 3.
ACHS-3-C04. Title: [Portrait of two girls in matching dresses]. Studio: Halvorsen Studio, Wren Gulch, Colo., from the printed imprint on the front of the mount. Date: approximately 1895 (1895~), basis: mount style and albumen print; “cabinet photographs” introduced in the United States in 1866 (TGM). People: Okerlund, Ida, and Okerlund, Marit, identified from the pencil inscription on the back, “Ida and Marit”; which girl is which is not recorded. Place: Wren Gulch (Colo.), from studio imprint. Format: Cabinet photographs (tgm001517); Albumen prints (tgm000227). Subjects: Portraits; Sisters, machine-suggested, reviewed and accepted. Inscription, back of mount, pencil: “Ida and Marit.”
ACHS-3-C09. Title: [Portrait of an unidentified man standing with a cane and a bowler hat]. Studio: none on the mount. Date: approximately 1880-1899 (1880~/1899~), basis: cabinet mount, albumen print, studio posing chair and painted backdrop of a type seen on other cards in the box. People: unidentified. The software grouped this card with ACHS-3-C11, a group portrait of pupils outside the Wren Gulch school, because the backdrop seems to match; a person compared the two, agreed the backdrop is probably the same, and recorded “backdrop possibly Halvorsen Studio, by comparison with ACHS-3-C11; machine grouping, reviewed.” The man remains unidentified. A face-match suggestion naming Walter Okerlund was rejected: there is no documented portrait of Walter to compare with, and the suggestion is kept in the history as rejected.
Checklist for each photograph
- Identifier matches the scan’s file name.
- Devised title in square brackets, with a distinguishing visible detail.
- Every inscription transcribed exactly, with its location and medium.
- Image and mount measured; process and format recorded with controlled terms.
- Date written in your one chosen form, with its basis.
- People recorded as identified, attributed or unidentified, each with a source.
- Place recorded only with its basis.
- Machine-suggested fields labeled until reviewed.
- Identification history kept, including rejected suggestions.
- Privacy and community concerns checked before the image goes online.
Next steps
Choose your date form and your vocabulary before the next box, and write them on one page your volunteers can keep beside them. Scan both sides of every mount, since the back is often where the evidence is; our photo and negative scanning page describes how prints and their mounts are captured, and the article on digitizing glass plate negatives covers the format that often has no mount to date at all. If you are weighing machine help for a large photograph collection, our AI-assisted cataloging service suggests fields that your staff review, and the historical societies page shows how the work fits a small organization.