Cursive reading guide

How to read cursive: a cursive reading worksheet

Q: Should I write down what I think a word says, or leave it blank?

Record it, marked. [unclear: Wednesday?] keeps your candidate available for the second pass and shows that you supplied it. An unmarked candidate becomes indistinguishable from a verified reading the next time somebody reads your transcript.

Reading cursive on a folded letter, a recipe card from a shoebox or a page from a journal can start with the same problem: the writing runs together and you cannot get a foothold. This page is a worksheet rather than a lesson. It gives you a sequence of things to inspect, write down and compare — record what is on the page, mark the units before you name any letter, inspect a set of letter pairs against the writer's own repeated shapes, transcribe a literal candidate, and mark everything you could not verify.

A cursive reading worksheet for family papers and personal documents. Last reviewed 15 August 2026.

If you arrived looking for how to read in cursive or for help reading in cursive, this guide uses both phrases to mean reading connected handwriting already on a page. It does not teach you to produce cursive writing.

Nothing here states that a step will work or that it will improve your reading. Each item is an observation to make and a note to write; what you conclude comes from your own page. The sequence needs no equipment beyond the document or an image of it, and you can run it beside machine output too: the cursive translator writes whatever text the service returns into an editable box, and this worksheet is how you check that text word by word against the image.

Step one: record what is in front of you

Before you name a single letter, write four answers at the top of your transcript. These are notes about your document, not conclusions about it — leave any of them blank or marked unknown rather than filling them in from impression.

Who wrote it, as far as you know? Write what you know and what you are assuming, separately. If you have another page you believe is in the same hand, put it beside this one and note which page it is.
Is there a date anywhere? Look for a dateline, a postmark, a printed letterhead, a stamp, a price, a reference to an event. Copy exactly what you find and where you found it. If there is nothing, write "undated" rather than estimating.
What kind of document is it? Note the parts you can see: a salutation and closing, a list of quantities, a column of figures, a repeated heading. List the words you can predict from that structure before you try to read them.
What is the physical state? Mark the fading, bleed-through, staining, folds crossing a line, and any later annotation in a different ink. Note which lines and words each one affects.

Then set the page up. Put light on it, or open the image; enlarge past life size so you can move around the page freely. If the strokes are faint, try greyscale and raised contrast on a copy, and try rotating the page so the writing slope runs vertically. Both are experiments: look at the result and keep whichever view you can work in. Note on your transcript if you read a word from an adjusted image rather than the original.

Step two: mark the units before you name any letter

Take one line and work through these four operations on it, writing down the result of each. Do not attempt to read the line while you are doing them.

Mark where the pen lifted

Draw a bracket at every point on the line where you judge the pen left the paper, using whatever you can see: the width of the gap, a break in the ink, whether a stroke arrives from the left. Count the units you have marked and write the number down. You now have a count to check any candidate reading against.

Record each unit's outline

For each bracketed unit, write down three things and nothing else: how many strokes rise above the body, how many drop below it, and the unit's length relative to its neighbours. Do this before naming a letter, so the outline is a record rather than a consequence of what you decided the word was.

Write a candidate for each unit, and where the candidate came from

Now propose readings. For each, note whether you got it from the outline, from the sentence around it, or from another occurrence on the page. Where two candidates both fit, write both. A unit with two live candidates is information; a unit silently reduced to one is not.

Test the candidates against the line as a whole

Read the line with your candidates in place and the unresolved units left blank, then read the sentence around it. Strike out any candidate that leaves the line ungrammatical, and note which ones you struck and why. Whatever survives goes into the transcript with the notation from step five — not as settled text.

Step three: six pairs, and what to inspect when you cannot decide

When a unit could be either of two letters, the table gives you something specific to inspect and record. It does not state how a writer forms a letter or which reading is more likely. Another instance attributed to the same writer in the same document is one comparison to record, not proof of a reading.

Six pairs, with what to inspect and how to record what you find. Not a description of any writer's hand.
Deciding between	What to inspect	What to compare it against	What to record
a / o	Where the stroke leaving the body towards the next letter starts: near the top of the body, or down at the baseline. Also whether the body is closed at the top.	A word on this page you have already confirmed contains one of the two — then look at what that instance does.	Both readings, the exit height you observed, and which confirmed word you compared against.
e / i	Whether the stroke encloses any white space inside itself, viewed at high magnification. Treat any dot as a separate observation, noted but not decisive.	An instance of each letter in a word you have already confirmed on this page.	Whether enclosed space was present, absent or undeterminable, and the confirmed word used.
n / u	Where the rounded part of the shape sits relative to the two verticals — at the top of the stroke, or at the bottom — and whether anything is added after it.	The same shape elsewhere in the same line, and in a word you have confirmed.	Where the curve sat, and both candidate readings of the whole word.
m / w	The number of peaks or valleys, their direction, and the height at which the exit stroke leaves the shape.	Read the word both ways, and also test it as nn and as uu. Then find a confirmed instance on the page.	The count you made, the exit height, and every reading that survived the test.
r / s	The foot of the letter where it meets the baseline: whether it closes or pinches, and whether the top is level or peaked.	An instance of the letter in a word on this page you have already read with confidence.	What the foot did, which word you compared against, and both readings if it stays open.
t / f	Whether anything descends below the baseline, the height of the ascender relative to other tall letters in the same word, and where any crossbar sits.	Tall letters elsewhere in the same line, so heights are compared within one hand rather than against a chart.	Presence or absence of a descender, the relative height, and whether you also considered l.

Two habits apply to every row. First, compare against another instance by the same writer in the same document rather than against a printed alphabet chart, and write down which instance you used — that is what makes your decision checkable later. Second, treat dots and crossbars as one observation among several: note where they are, note if they are missing, and do not let one of them close a question on its own.

Capitals: build a reference before you decide anything. Do not attempt a capital in isolation. Mark words whose position suggests a possible role — a signature, dateline or address — but treat that role as a hypothesis until the letters are supported by other evidence on the page. Record only a supported instance as your reference for that capital; otherwise write the candidates in brackets and leave it open.

Step four: build a key out of this document

Before working top to bottom, make a pass whose only purpose is to collect letterforms you can attach to a source. The key you end up with is a record of this document, and every entry in it cites the word it came from.

Start from possible roles, without treating them as readings

Position can suggest a possible role for a word — such as salutation, closing, signature, dateline, address or repeated heading — but does not identify its letters. List those hypotheses separately and note the visual evidence used to accept or reject each one.

Collect the repeated short words

Find the short words that recur across the page — the, and, of, to, in, is, was, you, we, for. Read one instance carefully, then look for the same outline elsewhere and check each candidate against the first. Where two instances of what you took for one word differ, write both down: that difference is data about this hand.

Write the key down with citations

Keep a list beside your transcript: each letter, and the word and line you established it from — a — from "and", line 3. When a doubtful stroke comes up later, compare it against your own confirmed instance rather than a general idea of the letter, and note in the transcript which key entry you used.

Use other pages, and record which page each entry came from

If you have more than one item you believe is in the same hand, start with whichever you find easiest to read and mark each key entry with its source page. Then check a sample of those entries against the harder page before relying on them: whether the two pages agree is something to verify, not assume. The same check applies within one document — compare the first line against the last.

Record habits as observations, with counts

Write down the shapes you observe and how many instances you saw: dots absent on a candidate i, crossbars falling past a letter, differing terminal and medial forms, or abbreviated months. The count records the size of your observation set; decide separately whether the examples support a later candidate reading.

Step five: transcribe line by line, with honest notation

One available convention is to keep one transcript line per source line and number the lines and pages. That creates a location reference for later image checks; it does not by itself prove that the transcription is complete.

The four conventions used in this worksheet separate candidate readings from supplied or unresolved material:

[unclear] — you can see writing but cannot read it. If you have a candidate, record it as [unclear: reading?] so the guess stays flagged rather than becoming fact.
[gap] — text is missing or unreadable for a physical reason: a tear, a fold, a blot, an ink loss, a trimmed edge, a stain. Add the cause and rough extent when you can: [gap: fold, approx. 4 words].
[expansion] — an abbreviation you have expanded, entered in square brackets only when you are confident: rec[eive]d. If you do not know what the abbreviation stands for, leave it exactly as written and note that it is unexpanded. An expansion you cannot cite is a reading you have supplied.
[sic] and a note — the writer's spelling, grammar or arithmetic is unusual but that is what the page says. Never correct it silently. Keep original spelling, original capitalisation and original punctuation throughout, including inconsistent spellings of the same name.

Mark inserted text, deletions and marginal notes too. A simple set is ^inserted^ for something written above the line, <struck> for something the writer crossed out but which you can still read, and [margin: ...] for a note running up the side.

Source: undated letter, family papers. Hand A.
Key: a, n, d confirmed from "and" (l.2); capital M from "May" (l.1).

01  [place name], May 12th
02  My dear [unclear: Emma?] —
03  Your letter came on Tuesday and we were all
04  very glad of it. The garden is [gap: blot, 1 word]
05  at last, though the rain has kept ^me^ from
06  the field these three days. Father sends his
07  <best> kind regards and asks after the
08  rec[eive]d parcel — did it come whole?
09  [margin: tell J. the seed is paid for]
10  Ever your affectionate [unclear] —

An illustration of the notation, not a real document. Every reading you supplied yourself stays inside brackets, so a later reader — including you — can tell evidence from inference.

Finish with a short header like the one above: where the document is, what it is, who wrote it if known, any stated date, and who transcribed it when. The header records the source identity claimed for this transcript; readers should still check it against the cited source.

Then make a second pass on another day, and go to the brackets first — they are the list of what you left open. Where a second pass changes a reading, record both the old and the new one and what made you change it, rather than overwriting the first.

Cursive reading practice on one line

Run this on a page you do not urgently need to read. Each step produces something written down; take as long as each one takes.

If you have no page to run it on, the cursive reading practice page supplies drill words and a ten-item test with its answer key.

Record the page whole page

Write the four answers from step one — writer, any date found and where, document type, physical state. Do not read a word of the body yet.
List candidate roles suggested by position whole page

Mark possible salutations, signatures, datelines and headings without treating position as proof of their letters. Start the key only from readings you can support with visible letterforms, and record the word and line used.
Bracket one line into units one line

Take a middle line and mark every point where you judge the pen lifted, ignoring meaning entirely. Write down the count.
Record the outlines same line

For each unit, note only strokes above the body, strokes below it, and relative length. Write a candidate for any unit you can, and where the candidate came from.
Inspect three doubtful letters three letters

Pick three and run the table: exit height for a/o, enclosed space for e/i, curve position for n/u and m/w, foot for r/s, descender for t/f. Record what you inspected and which confirmed instance you compared against.
Transcribe the line with notation same line

Write it out with [unclear], [gap] and any expansions bracketed. Leave every unresolved unit visibly unresolved.

Running this worksheet beside machine output

The left column is what this site's tool gives you, from its code. The right column is what only you can produce. Nothing here says which order to work in, or that either side improves the other.

What the tool gives you

One block of text in an editable box, from one request per image.
No confidence values, no highlighting and no marks: the page reads only the text field of the response.
This page displays only the response text field and does not render coordinates, columns, marginalia or form-field placement.
Text you can type into, so your own [unclear] and [?] marks go straight in.
A copy button and a download button, both acting on the box's current contents.
Checks: one accepted image up to 6 MB, an in-memory threshold of 10 requests per IP per minute, and a KV counter threshold of 100 per IP per UTC day.

What only you can do

Compare each returned word against the ink, and record which ones you verified.
Establish names, places and dates against the page and against an independent record.
Preserve original spelling, capitalisation, deletions and insertions, from the image.
Reconstruct anything that carried meaning by position on the page.
Decide what a wrong word would cost here, and set the depth of review from that.
Write down what you could not resolve, so it stays visible to the next reader.

A workable arrangement: submit the page at the cursive translator, put the returned text beside the image, and run this worksheet over it — bracket the units, inspect the doubtful letters against confirmed instances, and replace every word you have not verified with your own reading or with [unclear]. For notes and forms rather than correspondence, the handwriting to text page carries the same form with copy written for that material. For the request flow itself and a test protocol, see can AI read cursive?

Whatever arrives in the box is unmarked text. Nothing in it distinguishes a word that matches your page from one that does not, which is why your own brackets belong in the final transcript.

Questions about reading cursive

I never learned cursive. Can I still work through this?

The worksheet asks you to inspect shapes and write down what you see; it does not assume you were taught to write the script. If you want a document to start on, pick one whose content you can partly predict — a card, a recipe you know, a letter whose story you have heard — so you have something to test candidate readings against.

Where do I start on a page that looks like one continuous scribble?

Not at the first word. Start by marking positions that may contain a dateline, salutation or signature, but keep those as role hypotheses rather than letter identifications. Then take one middle line and mark where you judge the pen lifted, without trying to read it. Record both sets of observations for later comparison.

How do I decide between n and u when the writing is cramped?

Inspect where the rounded part of the shape sits relative to the two verticals, and record what you see. Then find the same shape in a word you have already confirmed on this page and compare the two directly. If it stays open, write both readings of the whole word in brackets and test each against the sentence.

The i's are not dotted and the t's are not crossed. Now what?

Work from the body of the letter rather than the mark. For i against e, inspect whether the stroke encloses white space. For t, check whether anything descends below the baseline and how the ascender compares with other tall letters in the same word. Note any crossbar you find further along the word as a separate observation rather than assigning it.

Should I write down what I think a word says, or leave it blank?

Record it, marked. [unclear: Wednesday?] keeps your candidate available for the second pass and shows that you supplied it. A blank discards work you did; an unmarked candidate becomes indistinguishable from a verified reading the next time somebody reads your transcript.

Should I expand abbreviations?

Only where you can establish what they stand for, and only inside square brackets so the original stays visible: rec[eive]d. Where you cannot, transcribe exactly as written and add a note saying it is unexpanded.

Is it worth photographing the page before transcribing?

Image software may provide enlargement, rotation and contrast controls on a copy. It is also the file you would submit if you use the tool. Record the capture conditions and note whenever a candidate reading came from an adjusted copy rather than the original view.

Can I just run it through the tool and skip all this?

You can submit the page, but what returns is unmarked text with no confidence values, so nothing in it tells you which words match your document. Whatever you intend to rely on still has to be checked against the image. Use the returned text as something to check, not as a result.

How long should a page take?

This page states no timings. Work until each unit is either verified against the image or marked as unresolved. Where a word will not resolve, bracket it as [unclear] and go on; you can return to it after the rest of the document has given you more confirmed letterforms.