Letter frequency analysis: why E wins and what the counts reveal

Count the letters in any long stretch of ordinary English and the same character keeps winning: E. Not by a nose, either — E shows up roughly once in every eight letters, while a letter like Q barely scrapes one in a thousand. That lopsidedness is not an accident of your particular paragraph. It is a deep, stable property of the language, and it has been used to crack codes, fingerprint languages, design keyboards, and quietly judge prose for centuries.

Letter frequency analysis is the simplest possible text measurement — tally each character, divide by the total — and yet the output reveals far more than the recipe suggests. This piece walks through where the famous ordering comes from, what the comparison against expected English actually tells you, and the real jobs this humble count still does.

ETAOIN SHRDLU: the canonical ordering

Printers and cryptanalysts long ago memorised the rough order of English letters by frequency: E, T, A, O, I, N, S, H, R and on down. The mnemonic "ETAOIN SHRDLU" comes from the layout of old Linotype machines, whose first two columns of keys were arranged in descending frequency order so a typesetter's hand travelled the shortest distance for the commonest letters.

The exact percentages depend on which corpus you measure — newspaper text, novels, and web writing all differ slightly — but the shape is remarkably stable. The letter frequency analyzer compares your text against a widely cited baseline drawn from large-scale corpus counts: E around 12.5%, T near 9.3%, A about 8.0%, with O, I, N, S, R, and H filling out the top tier. The bottom of the table — J, Q, X, Z — collectively accounts for well under 1% of typical English.

Why the distribution is so skewed

The unevenness is structural, not stylistic. English spelling encodes a few high-frequency phonemes with single common letters, and the most-used words in the language — "the", "of", "and", "to", "a", "in" — are themselves built from the high-frequency letters. E earns its crown partly through grammar: it ends countless words, marks the past tense in "-ed", appears in nearly every common suffix, and rides along in "the" itself, the single most frequent word in English. T and A get a similar boost from "the", "that", "and", and the article "a".

Because so much of the load is carried by a small set of function words, the letter distribution barely budges between authors. You can swap Hemingway for a software manual and E still wins. That stability is exactly what makes the count useful as a reference: deviations from it mean something.

How frequency analysis broke ciphers

The oldest serious use of letter counting is codebreaking. A substitution cipher replaces each letter with another, consistently — every E becomes, say, a K. The substitution hides which letter is which, but it cannot hide how often each symbol appears. So if you tally the ciphertext and one symbol dominates at around 12%, it is almost certainly standing in for E. The next most common is probably T or A, and from there you bootstrap the rest using common pairs and short words.

This technique, described by the 9th-century scholar Al-Kindi, is why simple substitution ciphers are useless for anything serious: the frequency fingerprint survives the encryption. Modern cryptography defeats it by destroying the one-to-one mapping — the same plaintext letter encrypts to different symbols each time — which flattens the count into noise. When you run ciphertext from a strong cipher through a frequency analyzer, every symbol sits near the same percentage. A flat distribution is itself a signal: it says "this isn't plain text."

Reading the expected column and the delta

The analyzer doesn't just count — it sets your text's percentages next to the expected English values and reports the delta (observed minus expected) for each letter. That comparison is where the tool earns its keep. A delta near zero across the board means your text behaves like generic English. Large deltas tell a story.

• A positive spike on an unusual letter — say Z or X running well above 0.1% — usually means a name, a technical term, or a foreign word is over-represented. Specialised vocabulary leaves a fingerprint here.
• A flattened top tier, where E and T sit closer to the middle of the pack, often signals that the sample is too short, is not English, or is something other than prose — a code, a list of part numbers, a table of data.
• A language that isn't English reshuffles the order entirely. French pushes E even higher and elevates accented forms; German lifts N and elevates compounds; Italian and Spanish raise the vowels A and O. Set the baseline aside and the raw ordering alone can hint at which language you're looking at.

What the options actually change

By default the tool is case-insensitive and counts letters only — digits and punctuation are excluded, so the percentages describe the alphabet itself rather than your formatting. That's the right default for comparing against the English baseline, because the baseline is a letters-only reference. Three toggles let you change the frame:

• Case sensitivity splits "E" and "e" into separate rows. Useful when you care about capitalisation patterns — acronym-heavy text, code, or title-case headings — but it makes the English-baseline comparison meaningless, since the baseline is lowercase.
• Include digits folds 0–9 into the count. Turn this on for serial numbers, dates, or data dumps where the numeric content is the point.
• Include punctuation adds symbols and marks to the tally. Helpful for spotting, for example, a passage that leans heavily on dashes or exclamation marks.

Whenever you include digits or punctuation, remember the percentages now describe all counted characters, so the letter figures will drop relative to the letters-only view. The expected-English column only makes sense in the default letters-only, lowercase mode.

Where it's still genuinely useful

Beyond codebreaking, letter counts do quiet work in several places. Word-game players use them to reason about tile draws and likely starting guesses — the high-frequency letters are the safe bets. Typeface and keyboard designers lean on the distribution so that the most-used letters get the most legible glyphs and the easiest reach. Language-detection systems use frequency profiles as a cheap first-pass classifier before anything heavier runs. And for writers, an oddly skewed table can flag an over-reliance on a particular name or jargon term that a plain read-through misses — the same instinct behind a word frequency analyzer, one level down at the character.

It pairs naturally with the broader text-analysis toolkit: where text entropy measures how predictable your symbol stream is in aggregate, letter frequency shows you the specific shape of the distribution underneath that number.

When you want to see the table for yourself — every letter's count, its share, the expected English value beside it, and the signed delta — run your text through the letter frequency analyzer. It charts the top letters, names your most frequent character, and lays the whole distribution next to the English baseline, so you can tell at a glance whether you're looking at ordinary prose, a cipher, another language, or a passage with a fingerprint all its own.