Reading Old Court Records: A Practical Guide

Reading old court records means solving three problems at once: a specialist legal script, a language that may not be the one you expect, and an abbreviation system that assumes you already know the formula. This guide sets out a working method — learn the template, build a letterform key from the clerk's own hand, and verify every name and number against the image. It is written for family historians who have hit a deed, will, or sessions roll they cannot get past.

Old court records are hard to read for three separate reasons at once: the script is a specialist legal hand, the language may not be the one you expect, and the text is compressed by an abbreviation system that assumes you already know the formula. The efficient way in is to work backwards from the formula. Court records are template documents — writs, pleadings, wills, presentments, and rolls reuse fixed clauses in a fixed order — so once you know the shape of the entry, you are only reading the variable slots: names, dates, sums, places, boundaries, witnesses. Learn the template first, build a letterform key from the clerk's own hand second, and treat every name and number as something to be verified against the image rather than inferred.

That is the method. The rest of this article works through each part of it.

What "court hand" actually is — and when it stops

Court hand (also curial or Chancery hand) is not simply "old handwriting." It is a family of formal documentary scripts developed for legal and administrative record-keeping, deliberately conservative, deliberately compressed, and often unlike the everyday writing of the same period. A clerk who wrote a fluent, legible personal letter in italic could produce a plea roll entry that looks almost cryptographic. Secretary hand — the workaday early-modern English hand you meet in wills, parish registers, and depositions — is a related script and much more commonly what a family historian actually faces. If your document is a will proved in 1610 or a Quarter Sessions examination, secretary hand is likely your problem, not plea-roll court hand.

There is a firm date worth knowing. The Proceedings in Courts of Justice Act, 4 George II c.26 required that from 25 March 1733 all writs, pleadings, indictments, records, judgments, recognizances, rolls and entries — including proceedings of courts leet, courts baron and customary courts — be in English "only, and not in Latin or French," written in a common legible hand "and not in any hand commonly called court hand, and in words at length and not abbreviated." Its stated scope covered courts of justice in England and the Court of Exchequer in Scotland, subject to stated exceptions.

That statute is the most useful orientation point for an English-records researcher. Before 1733, expect the possibility of Latin as the language of record, law French in certain functions, and court hand with heavy abbreviation. After 1733, expect English, spelled out. It does not make earlier records uniformly Latin, and it does not purge Latin tags from later ones — but it tells you which set of problems to prepare for.

The language question, handled properly

The most persistent misconception about old court records is that they are "all in Latin." They are not, and no universal percentage exists: the language of record varies by period, by jurisdiction, by court, and by document type within the same court. A pre-1733 English probate act clause may be Latin while the will it proves is English. A manorial roll may be Latin in its formulae and English in its marginal notes.

What survives across the boundary is a stock of Latin tags embedded in otherwise vernacular documents. These are the ones that actually recur:

versus / vs. — the opposing parties in a caption
scilicet / ss. — "to wit," introducing a particularization, very often in a venue line
videlicet / viz. — "namely," introducing particulars
et ux. — "and wife," in party descriptions; et al. — "and others"
ad respondendum — in a writ, directing response or appearance
fieri facias / fi. fa. — a writ of execution
habeas corpus — a command concerning custody
nisi prius — a trial arrangement
inter alia — "among other things"
L.S. / locus sigilli — the place where a seal is to be affixed, per the Society of American Archivists' definition. It marks a seal's location; it is not a signature.

Their presence reflects inherited legal drafting, not proof that the document is in Latin. If you are working through parish and probate material as well, the Latin formulae that actually appear in church records are a related but distinct vocabulary worth having alongside this one.

Keep three operations separate, always. Transcription records the visible text. Expansion makes an abbreviation convention explicit. Translation renders the language into modern English. Collapsing them is how errors become invisible: once you have written "to have and to hold" over a Latin habendum et tenendum, nobody downstream — including you, in six months — can tell what was on the page.

The formulas: what to read and what to skim

Formula-driven records work in your favour, because the boilerplate carries almost no information and the variable slots carry all of it. Learn one template per series and your reading speed changes.

Deeds and conveyances

A deed typically moves through: date and parties → recitals → authority and consideration → the operative grant and property description → habendum/tenendum ("to have and to hold") → restrictions or warranty → execution and testimonium. The variable slots are names, dates, consideration, capacities, the land itself, boundaries, and witnesses. Everything else is drafting convention. If you are working in land records specifically, the clause order and boilerplate of Anglo-American deeds repays a dedicated read, since the description units and metes-and-bounds language have their own failure modes.

Wills and probate

A will identifies the testator, disposes of property through bequests, and names executors and beneficiaries. In the record system it is followed by probate or administration — probate being, as Durham University's guide puts it, the process of proving a will and administering the estate of the deceased. The bequest sequence is where the genealogy lives: sons, daughters, wife, grandchildren, servants, and the residuary clause, often in an order that itself encodes family structure. The structure and hands of sixteenth- to eighteenth-century English wills are worth studying as a genre before you start on your own.

Manorial rolls

Manorial court rolls record the business over which the lord of the manor had jurisdiction — chiefly the court baron (chief tenants, internal regulation, land transfers by copy) and the court leet (peace-keeping jurisdiction, minor criminal offences). The National Archives' guide to manors and manorial records notes there is no set pattern for recording business, but that essoins (excuses) are typically listed first, followed by the names of the jury, then presentments. For a genealogist, copyhold land transfers on these rolls can be the only surviving evidence of a tenancy, since the "copy" entered on the roll was itself regarded as proof of title.

Quarter Sessions and equity

Quarter Sessions material comprises rolls, indictments, recognizances, papers, and examinations — the last frequently the most informative, because a witness examination can name relationships, ages, residences, and movements that no register records. Equity suits follow a different procedural sequence again; if your ancestor turns up in Chancery, the structure of bills, answers, and deposition bundles is the map you want before you open the file.

One caution. "Formulaic" does not mean uniform. Damaged pages, unusual litigation, local practice, clerk idiosyncrasy, mixed hands, and copied or abbreviated entries all break templates. Use the template as a hypothesis about what the page says, then check whether the page agrees.

The abbreviation system is a system

Abbreviations in legal records are not random shorthand. They fall into learnable categories, set out in Alpo Honkapohja's survey of manuscript abbreviations in Latin and English:

Suspension — omitting characters at the end of a word, often marked with a dot or stroke.
Contraction — omitting letters from the middle, frequently signalled by a macron above the word; often only the first and last letters survive (ds for deus, sm for secundum).
Brevigraphs and special signs — dedicated marks: the titulus/macron for an omitted nasal, the Tironian et, the p-stroke families, the -us and -rum signs, thorn.
Superscript letters — a raised letter standing in for a fuller ending.

The categories are stable; the individual signs are period- and language-sensitive. The same stroke does not mean the same thing in a 1450 plea roll and a 1620 English deed. Cappelli's dictionary remains the major reference for Latin and Italian forms, with the University of Zurich's Ad Fontes abbreviation resources as the practical way in. For a broader treatment of the marks themselves, see the guide to manuscript abbreviations and ligatures.

The discipline that matters: expand transparently. Write `yo[u]r`, not `your`. Bracket what you supplied. If you cannot resolve a mark, leave it and flag it rather than guessing a plausible ending.

A working method for a page you cannot read

Establish the series and the date range. Court and government records travel in series with house conventions; knowing which series you are in tells you the likely language, the likely hand, and the likely template. The wider question of how court and government record series are transcribed and indexed for public access is the context your single page sits inside.
Find a parallel entry you can already read. In a formula-driven series, the entry above or below yours is very nearly identical. Read the easy one first and use it as a key.
Build an alphabet from this clerk. Not from a textbook. Collect each letterform from several lines of the same hand, including its variants in initial, medial, and final position. The National Archives' tutorials on reading old documents and Cambridge's English Handwriting 1500–1700 course — with its alphabets, transcription conventions, and sample transcriptions — are the two resources to work through if you do this often. The general procedure for approaching a hand you have never seen before applies here as much as anywhere.
Transcribe the boilerplate, then attack the slots. Do the easy clauses to warm up and confirm your letterform key. Then slow down on names, dates, sums, and boundaries.
Mark uncertainty explicitly. Square brackets for supplied text, a bracketed question mark for a doubtful reading, a note for illegible extent. A transcription that records its own doubt is more useful than one that hides it.
Verify the high-stakes tokens against the image. Names, numbers, negations, legal scope, and any abbreviation you expanded. These are where errors do real damage in a family tree.

The specific traps are visual, not contextual: long s read as f, the secretary e, r, and c, minim strings where m, n, i, u and v dissolve into identical strokes, thorn mistaken for y, and abbreviation signs read as letters. A word that makes perfect contextual sense can still be the wrong word. Context proposes; the image decides.

Where machine transcription helps on this material — and where it doesn't

At some point you will have more pages than hours, and that is where machine transcription earns its place. Be clear-eyed about what the evidence supports. Handwritten text recognition figures are dataset-specific, never universal: a 2025 comparative evaluation of HTR engines recorded a 1.50% character error rate on its Roman-type dataset, but court hand was not directly benchmarked there, and low error on printed Roman type transfers nowhere near dense abbreviated legal manuscript. Vendor claims about public court-hand models — including specialist ones such as Transkribus's Latin Court Hand plea-roll model — are capability claims until you test them on your own series.

General chatbots are the wrong instrument here, for a reason specific to legal records. An evaluation of a multimodal LLM on ICDAR historical corpora found that where inaccuracy was extreme, the model generally produced text unrelated to the underlying image — hallucination rather than misreading. That study is one model on one set of corpora, and it also found many perfectly transcribed samples, so it is not a verdict on every general model; but the failure mode it names is exactly the one that formula-driven records invite. A model that has absorbed thousands of deeds knows what a habendum clause usually says. Asked to read a damaged one, it can produce the usual clause fluently, in the right register, with your ancestor's name plausibly inserted. Garbled output announces itself. Fluent output does not. The same dynamic is examined in more detail in the discussion of why fluent AI transcriptions are the dangerous kind.

This is the stage where Leo fits. Its transcription model, ATR-1, is purpose-built for Latin-script manuscript and printed material of roughly the past five hundred years — the alphabet is the constraint, not the language, so English wills, pre-1733 Latin-language pleadings, and law French all fall inside it, while Greek, Cyrillic, Hebrew and Arabic scripts do not. Two things matter for court records specifically. First, it is trained to transcribe what is on the page rather than to normalize it: archaic orthography survives, strikethroughs and marginal additions survive, and an abbreviation mark is not silently resolved into the modern expansion it probably stands for — which is precisely the judgment you want to make yourself, in brackets. Second, there is no per-series model training step, so you can test it against a page you have already transcribed by hand before trusting it on a hundred you haven't. Leo also runs translation as a separate operation that writes to a new tab, leaving the base transcription intact — transcription and translation stay two jobs, as they should. One honest limitation, relevant here: pages dominated by pre-printed structure with handwriting filled in — printed deed-book and register forms — are the known weak spot, where the model can favour the printed headers over the manuscript entries.

Whatever tool you use, the boundary does not move: a machine draft is a first pass, and names, dates, sums and boundaries get checked against the image.

What actually accumulates

The genuine skill in reading old court records is not decipherment of individual letters. It is pattern knowledge — knowing that a Quarter Sessions recognizance has a shape, that a manorial presentment follows the jury list, that ss. means a venue line is coming, that the third clause of the will is where the daughters appear. That knowledge compounds. The tenth entry in a series takes a fraction of the time the first one did, and by the fiftieth you are reading the variable slots almost directly, the way the clerk expected his colleagues to.

So spend the effort on the series, not the page. Photograph or note the parallel entries, keep a running letterform key for each clerk's hand, and record your uncertainties honestly enough that you can revisit them when a later document resolves them. A transcription with three bracketed doubts and a clear note of what you could not read is a document you can build a family tree on. A clean one that quietly guessed is not.

Frequently Asked Questions

How do I start reading old court records when I can't make out the handwriting?

Start with the formula, not the letters. Court records are template documents: writs, deeds, wills, presentments and rolls reuse fixed clauses in a fixed order, so once you know the shape of the entry you are only really reading the variable slots — names, dates, sums, places, boundaries, witnesses. Establish the series and date range, find a parallel entry above or below yours that you can already read, and use it as a key. Then build a letterform alphabet from that clerk's own hand rather than a textbook, transcribe the boilerplate to warm up, and slow down on the slots.

Are old court records all written in Latin?

No. That is the most persistent misconception about them, and no universal percentage exists: the language of record varies by period, jurisdiction, court, and even by document type within the same court. A pre-1733 English probate act clause may be Latin while the will it proves is English; a manorial roll may be Latin in its formulae and English in its marginal notes. What does survive across the boundary is a stock of Latin tags — versus, scilicet (ss.), videlicet (viz.), et ux., fieri facias, nisi prius, locus sigilli — embedded in otherwise vernacular documents. Their presence reflects inherited legal drafting, not the document's language.

When did English court records stop using Latin and court hand?

From 25 March 1733, under the Proceedings in Courts of Justice Act, 4 George II c.26. The statute required that all writs, pleadings, indictments, records, judgments, recognizances, rolls and entries — including proceedings of courts leet, courts baron and customary courts — be in English "only, and not in Latin or French," written in a common legible hand "and not in any hand commonly called court hand, and in words at length and not abbreviated." Its stated scope covered courts of justice in England and the Court of Exchequer in Scotland, subject to exceptions. It did not make earlier records uniformly Latin, nor purge Latin tags from later ones.

What is the difference between court hand and secretary hand?

Court hand (also curial or Chancery hand) is a family of formal documentary scripts developed for legal and administrative record-keeping — deliberately conservative, heavily compressed, and often unlike the everyday writing of the same period. Secretary hand is the related workaday early-modern English hand found in wills, parish registers and depositions, and it is far more commonly what a family historian actually faces. A clerk capable of a fluent, legible italic letter could still produce a plea roll entry that looks almost cryptographic. If your document is a will proved in 1610 or a Quarter Sessions examination, secretary hand is likely your problem.

Can AI transcribe old court records reliably?

It can produce a useful first pass, but the draft still has to be checked against the image. General chatbots are the wrong instrument for legal records specifically: where inaccuracy became extreme in one published evaluation, the model produced text unrelated to the underlying image — hallucination rather than misreading. Formula-driven records invite exactly that. A model that has absorbed thousands of deeds knows what a habendum clause usually says, and can render a damaged one fluently with your ancestor's name plausibly inserted. Garbled output announces itself; fluent output does not. Names, dates, sums and boundaries always get verified manually.