Scrabble letter distribution

There are 100 tiles in the bag — 98 letters and 2 blanks — worth 187 points in total. Here is how many of each letter there are, and the exact chance of seeing one in an opening rack of 7.

The bag is not an alphabet. E alone accounts for 12 tiles, while J, K, Q, X, Z have one each. The five vowels take up 42 of the 98 letter tiles — which is why a rack full of consonants feels wrong, and is.
TileIn the bagPointsIn opening rack
E12160.3%
A9149.4%
I9149.4%
O8145.3%
N6136.1%
R6136.1%
T6136.1%
D4225.5%
L4125.5%
S4125.5%
U4125.5%
G3219.7%
Blank2013.6%
B2313.6%
C2313.6%
F2413.6%
H2413.6%
M2313.6%
P2313.6%
V2413.6%
W2413.6%
Y2413.6%
J187.0%
K157.0%
Q1107.0%
X187.0%
Z1107.0%

“In opening rack” is the chance at least one turns up when you draw 7 tiles from a full bag — exact, not simulated. Every single-copy tile sits at exactly 7%, because drawing 7 of 100 gives any one specific tile that chance.

Which Scrabble is this? The English-language set. Tile distributions differ by language — French and German sets have 102 tiles, Italian 120 — but they do not differ between the two English word lists: NWL in North America and Collins everywhere else disagree about which words are good, never about what is in the bag.

And why “opening” rack? Every figure here assumes a full bag, which is true exactly once a game. What breaks it isn’t the turn number though, it’s information: these are the odds before you know anything, so they hold for any draw you make with your eyes shut. From your second turn on, the bag is whatever is left, your own rack and the board already rule tiles out, and a player tracking the bag is working from a shorter list than this one.

Where these numbers came from

Alfred M. Butts worked the tile counts out during the Depression, and the set has barely moved since. The most-repeated version of that story is that he fixed the counts by tallying letters on the front page of the New York Times — which is on a hundred sites and in none of the four institutional sources below, each checked for it. So rather than repeat it, here is the part that can be tested.

Tiles vs. letter frequency
0.90
Measured across
1.57M letters

That is the correlation between how many tiles a letter gets and how often it actually turns up, counted over every letter of every word in ENABLE — 1,570,540 of them across 172,823 words. Whatever he counted, he was not guessing.

Where he overruled the frequency

The gaps are more interesting than the fit, and one is not close. The S gets 4 tiles — 4.1% of the letter tiles against 9.5% of the letters, a shortfall of 5.4 points and 2.7× the next biggest gap (the C). Under-tiling the S is what stops pluralising everything in sight from being the whole game, and it is the reason an S is the tile people hoard.

One caveat, because it cuts in the same direction: a word list is not running text. A dictionary counts every plural as its own entry, which inflates the S against ordinary prose, so the true shortfall is smaller than 5.4 points — but it is a shortfall on any measure, and it is still the biggest one in the set.

The tiles nobody can place, and the words that save them

The counts are only half of what makes a tile awkward — the other half is whether there is anywhere to put it. Four letters appear in no two-letter word at all in ENABLE: C, Q, V, Z. Two of those are rescued at a rated game by QI and ZA, which ENABLE does not carry; a C only by Collins’ CH. The V is stuck in every list there is.

The genuinely odd one is XU, a Vietnamese coin, which is in ENABLE and is the reason an X — worth 8 and with only 1 in the bag — is far less of a problem tile than a V worth 4. Face value is not difficulty.

What is actually documented

Four separate sources rather than four pages repeating one: the manufacturer for what is in the box, a museum for the naming history, an encyclopaedia for the dates, and MIT’s invention programme for the design principle. Repeated widely, and carried by none of the four sources above, each of which was checked for it. It may well be true; it is not something this page can stand behind, so it tests the claim instead of restating it.

Try next

Odds The Law uses cookies for analytics and to show ads. Decline and neither will load. See our Privacy Policy for details.