Should a Random-Letter Wheel Give Every Letter an Equal Slice, or Match Real English Letter Frequency?
A random-letter wheel built with 26 equal slices gives every letter exactly 1 ÷ 26 ≈ 3.85%, a genuinely unbiased pick appropriate whenever every letter needs an equal shot — a hangman starting-letter draw, an alphabet-order lottery, or any game where fairness among the 26 options is the actual goal. Weighting a wheel to match real English letter frequency instead gives E about 12.7% and Z about 0.074%, a gap of roughly 172-to-1 between the two, useful only when the wheel is meant to simulate realistic English text rather than to pick fairly among 26 equally valid letters.
A worked example with numbers
The gap between equal-slice and frequency-weighted odds is large across most of the alphabet, not just at its extremes. Against an equal 3.85% baseline, E’s real 12.7% share is about 3.3 times larger, T’s 9.1% is about 2.37 times larger, A’s 8.2% is about 2.13 times larger, and O’s 7.5% is about 1.95 times larger — every one of the four most common English letters individually beats its equal-slice share by roughly double or more. At the other end, J at 0.16% and Q at 0.12% sit at only about 4% and 3% of the equal-slice share respectively, and Z at 0.074% sits at just 2% of it — meaning a frequency-weighted wheel gives Z a slice roughly 50 times narrower than the equal-slice version would, on top of the 172-to-1 gap against E specifically. The full frequency table, most commonly attributed to Robert Lewand’s corpus-based analysis, runs across every letter from A at 8.2% down to Z at 0.074%, and none of it changes if the wheel happens to use a different visual style — the underlying probabilities are a property of the weighting choice, not the wheel’s appearance.
How to set the rule before the result
Building the equal-slice version needs no special configuration: the random letter generator draws uniformly by design, with no extra step required. Building the frequency-weighted version means entering all 26 letters into a weighted random picker or the custom weighted wheel with their real-world frequency numbers as weights, converting each percentage to a slice proportional to its share of the total — the same weight-to-probability mapping process used for any weighted random choice on this site, just applied to 26 specific numbers instead of a handful of custom ones. To confirm either version is actually working as configured, log a long run of draws and check that the observed share for each letter clusters near its intended target, the same verification approach used to test any weighted or unweighted picker.
Common mistakes that change the odds or the process
A common mistake is assuming frequency-weighting is somehow “more random” or “more fair” than equal slices; it is neither — both describe precisely defined probability distributions, and frequency-weighting is deliberately less fair among individual letters if the actual goal is equal opportunity, meaning a task like assigning students to teams based on their initial letter should never use frequency-weighting, since it would systematically favor students whose names start with common letters. A second mistake is applying an English-language frequency table to a wheel meant for a different language; French, German, and Spanish letter frequencies differ meaningfully from English’s — a French-text simulator using English’s 12.7% for E would misrepresent French, where other letters carry more weight. A third mistake is treating the classic frequency table as a universal constant rather than a summary of general prose; frequencies drawn from URLs, computer code, or license-plate combinations can differ substantially from ordinary written English, so what counts as “realistic” always depends on what kind of text the wheel is meant to imitate.
Where this method stops being appropriate
This comparison covers drawing a single letter with replacement, where every draw is independent of the last; it does not describe drawing several letters in a row from a fixed, shrinking pool without replacement, a different system covered in the second case below. Frequency tables are also a statistical summary across large text corpora and will never exactly predict the letter composition of any one specific short word or sentence — they describe averages over long text, not a guarantee about any individual sample.
How the random source fits into the rule
Wikipedia’s letter frequency reference documents the full 26-letter English frequency table used above, commonly attributed to Robert Lewand’s corpus-based analysis; Wikipedia’s Scrabble letter distributions reference documents the fixed 100-tile Scrabble letter distribution referenced as a third, real-world designed system.
The full 26-letter frequency table against the equal-slice baseline
| Letter | English frequency | Letter | English frequency |
|---|---|---|---|
| E | 12.7% | M | 2.4% |
| T | 9.1% | W | 2.4% |
| A | 8.2% | F | 2.2% |
| O | 7.5% | G | 2.0% |
| I | 7.0% | Y | 2.0% |
| N | 6.7% | P | 1.9% |
| S | 6.3% | B | 1.5% |
| H | 6.1% | V | 0.98% |
| R | 6.0% | K | 0.77% |
| D | 4.3% | J | 0.16% |
| L | 4.0% | X | 0.15% |
| C | 2.8% | Q | 0.12% |
| U | 2.8% | Z | 0.074% |
The equal-slice baseline for every letter is a flat 3.846%; every row above or below that figure marks how much a frequency-weighted wheel would favor or penalize that specific letter relative to a plain alphabetical draw.
A second case: Scrabble’s tile bag is neither equal nor purely frequency-weighted
Scrabble’s 100-tile bag holds a fixed integer count per letter — 12 E, 9 A, 9 I, 8 O, and down to 1 each for J, K, Q, X, and Z, plus 2 blanks — a distribution built from period-newspaper letter-frequency analysis but expressed as whole tiles rather than exact percentages, and crucially drawn without replacement. That last detail changes the odds as a game proceeds: the first tile drawn has a 12/100 = 12% chance of being E, but once one E has already been drawn, the next tile’s chance of being E drops to 11/99 ≈ 11.11% — a shrinking-population effect a wheel, which always draws with replacement, does not share. A frequency-weighted wheel can approximate Scrabble’s starting odds reasonably well, but it cannot reproduce this narrowing effect without being redesigned as a without-replacement draw instead.
Choosing between equal and frequency-weighted for a specific task
Equal slices are the objectively correct choice whenever the task requires fairness among the 26 letters themselves — assigning turn order, picking a genuinely arbitrary starting letter for a word game, or any situation where favoring E over Z would be considered a flaw rather than a feature. Frequency-weighting is the objectively correct choice whenever the task is instead trying to imitate realistic English text — generating filler text for testing a layout, producing a more natural-feeling word-guessing game, or building practice material for a cipher-breaking exercise that specifically relies on real language’s letter-frequency patterns. Picking the wrong one for a given task does not make the wheel less random in either direction; it makes it randomly wrong for the specific job it was built to do.
Vowels vs. consonants: what the frequency table adds up to
Summing the five vowel letters’ real English frequencies — A at 8.2%, E at 12.7%, I at 7.0%, O at 7.5%, and U at 2.8% — gives a combined 38.2%, against 61.8% for the 21 consonants together. An equal-slice wheel, by contrast, always gives the vowel group a flat 5 ÷ 26 ≈ 19.2% combined share and the consonant group the remaining 80.8%, regardless of how English text actually behaves. That roughly two-to-one gap between 19.2% and 38.2% is a useful quick check for telling the two wheel types apart at a glance: a frequency-weighted wheel should visibly favor its five vowel slices as a group, while an equal-slice wheel gives every one of the 26 slices, vowel or consonant, the identical width.