Fraktur OCR: How to Read and Convert Blackletter Print

Fraktur is the angular German blackletter type that filled books, Bibles and newspapers for four centuries. A person can learn to read it in an afternoon, but OCR software has long found it hard. This guide explains what Fraktur is, the letters that trip both readers and machines, which tools to use now, and how to read it yourself.

If you have opened a German book printed before the Second World War, a nineteenth-century newspaper from Vienna or Milwaukee, or a city directory that lists your ancestor, you have probably met Fraktur. It looks forbidding, but the type is regular: once you know a dozen tricky letters, you can read it. Machines have a different problem, because OCR engines were built for modern roman type. Fraktur OCR now sits close to handwritten text recognition, because the newest tools read historical print and handwriting in the same way.

What Fraktur is: blackletter, Textura, Schwabacher and Fraktur

Blackletter is the family name for the dense, broken-stroke scripts of the late Middle Ages and the type cut to imitate them. Three blackletter typefaces matter to anyone reading old German print:

Textura (or textualis) is the tallest and most rigid, a picket fence of straight strokes broken at the top and bottom. Gutenberg's 42-line Bible was set in textualis, but it was rarely used for type after the early years.
Schwabacher is rounder and closer to handwriting, with a distinctive lowercase g and capital H. It was the usual German type of the late fifteenth century and of the first Luther Bibles. Fraktur displaced it as the main German type during the sixteenth century, but printers kept Schwabacher for emphasis, the way English printers use italic.
Fraktur first appeared in the early sixteenth century in work commissioned by Emperor Maximilian I, including his prayer book of 1513. Its lowercase o is a quick test: the left side of the bowl is broken into an angle, while the right side stays curved.

German printing used Fraktur for far longer than the rest of Europe used blackletter. In the nineteenth century scholarly books moved to roman type (Antiqua), but most literature and newspapers stayed in Fraktur. It ended by decree: an edict of 3 January 1941 banned blackletter type, and a second order soon banned Kurrent handwriting in schools as well. Fraktur still appears after that date in reprints, signs and mastheads.

A naming trap. Libraries and OCR evaluations often call blackletter "Gothic". In those reports, "Gothic" means Fraktur and its relatives, not the sans-serif type that typographers also call Gothic.

Fraktur is print; Kurrent is handwriting

Fraktur is often confused with Kurrent, the German handwriting of the same centuries, and with Sütterlin, its early twentieth-century school form. Fraktur is type, identical every time it appears; Kurrent is a pen hand, different for every writer. Reading Kurrent is a paleography problem, and our guide on how to read German Kurrent covers it. A church register entry in ink is Kurrent; the printed column headings above it are often Fraktur.

In America, Pennsylvania German Fraktur means something else: decorated documents such as baptismal certificates (Taufscheine), made mostly between 1740 and 1860, at first entirely by hand and later with printed text.

Where you will meet Fraktur

Books, Bibles, hymnals and catechisms printed in German-speaking Europe from the sixteenth century to the 1940s
Newspapers and periodicals. ANNO, the Austrian National Library's newspaper portal, lists more than 1,600 titles and about 28 million pages (October 2026), much of it in Fraktur.
City directories and address books, a staple of German genealogy, which list householders by street and trade
Official print: government gazettes, laws and public notices
The German-language press abroad, including the German newspapers of nineteenth-century America
Church books and certificates with printed headings or text, where the entries themselves are handwritten
Other languages. Danish, Swedish and Latvian printers used Fraktur too, so a Scandinavian or Baltic source may need the same skills.

The letters that trip readers and OCR

Most Fraktur letters are close enough to roman type to guess. The few that are not cause most OCR errors too.

Long s and f. Fraktur has two lowercase s forms. The long ſ is used at the start and in the middle of a syllable, and the round s at the end of a word or of each part of a compound, under the spelling rules of 1901: das Haus but im Hauſe. The long ſ looks almost exactly like f, and both drop below the line. The difference is the crossbar: f has a stroke that crosses the stem on both sides, while ſ has at most a nub on the left. An engine that has not learned ſ reads geſtorben (died) as "geftorben".

ß. The sharp s is built from a long ſ and a tailed z (ʒ), and in Fraktur you can still see the two halves.

k and t, c and e, n and u. k rises well above the other letters, with a loop at the top right, while t barely clears them. e closes into a small eye where c stays open, and n is joined at the top where u is joined at the bottom. On worn type these differences vanish, and a dictionary-based corrector may then "fix" the result into the wrong word.

r and r rotunda. After a letter with a round right side, such as o, b, p or the Fraktur d, many typefaces use a second r shaped like the numeral 2 with a tail, the r rotunda (ꝛ). Readers take it for a z or a 2.

Ligatures: ch, ck, tz, ſt. Fraktur casts these pairs as single pieces of type. When a word is letter-spaced for emphasis (Sperrsatz, Fraktur's italics), they stay together while every other letter is spaced apart, and a word like "G e ſ ch i ch t e" can come out of OCR as scattered letters.

Umlauts. The two dots began as a small e written above the vowel, and older Fraktur prints often show exactly that: aͤ, oͤ, uͤ. Later type uses two dots or two short strokes, and capital umlauts were often set out in full: Ueber, Aerzte, Oel. OCR may drop the mark as noise.

Capitals. A, U and N are easily confused, as are B and V, C and E, and M and W. Most older Fraktur uses one shape for both I and J, which looks more like a J.

Mixed type. Fraktur text often switches to roman type for Latin or French words, names and foreign titles, and to Schwabacher for emphasis. Pages that mix typefaces are among the hardest for OCR.

A Fraktur alphabet: the letters worth learning first

The list below covers the forms that differ most from modern type.

ſ (long s) — What to look for: A tall stem dropping below the line, with no crossbar or only a nub on the left; Often misread as: f
f — What to look for: Like ſ, but with a stroke crossing the stem on both sides; Often misread as: ſ
s (round) — What to look for: Small and closed, much shorter than ſ; only at the end of a word or word part; Often misread as: o, on worn type
ß — What to look for: A long ſ joined to a tailed z; Often misread as: B, ſs
k — What to look for: Tall, with a loop at the top right and a short arm; Often misread as: t, f
t — What to look for: Short, barely above the other lowercase letters, with a crossbar; Often misread as: k, c
c / e — What to look for: Same body, but e closes into a small eye at the top; Often misread as: each other
n / u — What to look for: Two straight strokes joined at the top (n) or at the bottom (u); Often misread as: each other
ꝛ (r rotunda) — What to look for: A 2-shaped r, after o, b, p, d and similar letters; Often misread as: z, 2
ʒ (z) — What to look for: A z with a tail curling below the line; Often misread as: 3, h
h — What to look for: A tall stroke whose right leg drops below the line; Often misread as: b, y
A / U / N — What to look for: Similar broken capitals; check the top of the letter and the strokes inside; Often misread as: each other
B / V — What to look for: Rounded capitals; compare with a word you know; Often misread as: each other
I / J — What to look for: One shape for both in most older Fraktur; Often misread as: either
ch, ck, tz, ſt — What to look for: Joined pairs that stay together when the word is spaced out; Often misread as: separate letters
aͤ, oͤ, uͤ — What to look for: A small e above the vowel, the older form of ä, ö, ü; Often misread as: a, o, u plus noise

Typefaces differ between printers, so treat any chart as a starting point and check a doubtful letter against a word you are sure of.

How OCR has handled Fraktur

Conventional OCR engines match each letter against the shapes they were trained on, mostly modern roman type, and then often run the result through a modern dictionary that "corrects" historical spellings into errors. The Europeana Newspapers evaluation (Pletschacher et al., 2015), run with ABBYY FineReader Engine 11 on 600 historical newspaper pages, found that the engine recognised 81.4% of words on pages set in roman type, against 67.3% on blackletter ("Gothic") pages and 64.0% on pages that mixed the two. That measure ignores word order, so the readable text was worse still.

Specialised models narrowed the gap. Tesseract, the open-source engine, has a dedicated German Fraktur model (deu_latf, formerly frk). OCR4all, a free, open-source workflow built for historical print, lets you correct output and train a model on your own book; its authors reported that it could outperform ABBYY FineReader on nineteenth-century novels (Reul et al., 2019) when suitable pretrained models were available. These tools work well on clean pages but take some setup, and a model trained on one printer's type may need retraining for the next.

The newest option is AI models that read the whole image in context. A 2025 study of German city directories printed between 1754 and 1870 (Greif et al., 2025) found that the best multimodal model significantly outperformed conventional OCR, including Tesseract's Fraktur model. Reading words in context, they rarely confuse ſ with f. But general-purpose models tend to quietly modernise spelling, turning Theil into Teil, which matters if you need the text as printed. Our guide to HTR vs OCR explains how the two families of tools differ, and OCR for old books covers historical print more widely.

Which tool to use now

A few pages you need to understand. Learn the alphabet above and read them yourself.
A digitised newspaper or book from a library portal. Check whether the library already offers searchable text. If it does, search it with variant spellings and both s forms, since older OCR may have read ſ as f.
A whole Fraktur book or run of newspapers, and you are comfortable with technical tools. Tesseract's Fraktur model or OCR4all will give a good first text on clean pages.
Historical text you will quote, cite or transcribe for a family history. Use a model built for historical documents that keeps the spelling as printed, then check it against the image.

Scan well first, because Fraktur's hairlines vanish at low resolution: see what DPI to scan old documents for settings. Then test any tool on your own pages before you commit a whole collection to it, because typefaces and printing vary too much for any general result to predict yours.

How to read Fraktur yourself

Start with words you know. Names, places, dates and common words such as und, der, die, das and ſind give you letter forms you are sure of.
Learn the six troublemakers first: ſ and f, k and t, the r rotunda, and the capitals. Most other letters are close enough to roman type to guess.
Read whole words, not letters. Fraktur is regular type, so once a word's outline is familiar, you will recognise it on sight.
Expect old spelling. Before the 1901 reforms, German spelled many words differently: Theil, Thür, Noth, ſeyn. Do not "correct" them.
Keep a key of your own. Note each puzzling letter with a word where you found it.

Writing the transcription: keep what is printed, or record what you changed

Decide how you will handle the forms modern German lacks, and state your choices at the top of the transcription.

Long s. Keep ſ for a faithful transcription, or write s and say so. Never mix the two.
R rotunda. Most editors write a plain r; keep ꝛ only if the typography itself is your subject.
Umlauts. Write aͤ for the superscript-e form, or ä, and note which you chose.
Ligatures. Write the letters separately. A ligature is the printer's choice, not a spelling.
Spelling. Keep it as printed. A normalised version is useful for searching and reading, but it is a separate layer, and historical language normalization explains how to make one without losing the original.

Where Leo fits

A page or two of Fraktur is an afternoon's work. A run of newspapers or a city directory is a project, and a first machine transcription gets you to the reading much faster.

Leo's transcription model, ATR-1, is specially tuned for historical documents, printed as well as handwritten: the kind of material conventional OCR was never designed for. It is trained to transcribe what is on the page rather than normalise it, so the long ſ stays a long ſ, Theil stays Theil, and nothing is quietly modernised the way general-purpose AI models tend to do. ATR-1 reads the image at high resolution, where general models downsample it, and needs no training on your typeface before you start. Built-in failure detection hides output that looks wrong, retries it, and refunds the credit if it cannot succeed. The errors that do get through are the recoverable kind: a wrong letter that you check against the image shown beside the text.

Leo also has a model library: one menu in the transcription window, with ATR-1 as the default at 1 credit an image. For clean modern print, such as a twentieth-century book in roman type or a typescript, a small model such as GPT-6 Luna reads an image for 0.02 credits. Fraktur is historical print, so send it to ATR-1, not to the cheap model. In our tests, small models did well on clean modern books and typescripts, less well on Fraktur, and tended to modernise older spelling.

The workflow suits a German family archive:

Upload scans, PDFs or photographs straight from your phone
Compare readings. Run one of the major general models over the same image into its own tab, while ATR-1's transcription stays the default.
Translate a German transcription into English in a separate tab, so the German text is never overwritten
Search every transcription at once, with fuzzy search to catch variant spellings of a surname
Export to Word, PDF, HTML or TEI, or share a read-only link

Whatever produces the first draft, check names, dates and numbers against the image. Our method for verifying transcription accuracy shows how. Every correction you make in Leo feeds back into training, so later readings improve.

Leo has a free plan, so you can upload a few images of Fraktur and see the result for yourself.

Frequently Asked Questions

Can OCR read Fraktur?

Yes, with the right tools. Engines built for modern roman type recognise markedly fewer words on blackletter pages. Specialised Fraktur models, such as Tesseract's German Fraktur model and OCR4all, do much better on clean pages, and AI models that read the whole image in context have done better still in recent research. Check any result against the page, especially names and numbers.

How do you tell the long s from f in Fraktur?

Look at the crossbar. The letter f has a stroke that crosses the stem on both sides, while the long ſ has no crossbar or only a small nub on the left. The position helps too: German uses ſ at the start and in the middle of syllables, and the round s at the end of a word or word part, so a long letter at the end of a word is almost always f.

When did Germany stop using Fraktur?

Scholarly books moved to roman type during the nineteenth century, but most literature and newspapers stayed in Fraktur until an edict of 3 January 1941 banned blackletter type. Fraktur still appears after 1941 in reprints, signs and mastheads.

Is Fraktur the same as Gothic type?

Fraktur is one kind of blackletter, the family that libraries and OCR evaluations often call "Gothic". Its relatives are the older Textura and Schwabacher. Fraktur became the standard German type in the sixteenth century, and stayed in use far longer than blackletter did anywhere else.

Can AI read Fraktur?

Yes, and AI models that read the whole image in context have outperformed conventional OCR on Fraktur in recent studies. General-purpose models tend to quietly modernise the spelling, though, which loses the text as printed. Leo's ATR-1 is specially tuned for historical documents, printed and handwritten, and trained to keep what is on the page, long s included. It reads the image at high resolution, and output that looks wrong is hidden, retried and refunded if it cannot succeed. Check the transcription against the image beside it.