10 worst hands

10 worst hands covers 10 hands. Their average equity against a single random opponent is 34.9%, from 32o at 32.3% (rank 169 of 169) down to 73o at 36.6% (rank 160). That is a spread of -4.3 points inside a single hand type.

At a nine-handed table (eight opponents) the average falls to 7%, and even 32o holds only 6.2%. The all-hand average against eight opponents is 12.2%, so this type sits below the field once the pot goes multiway. 32o ranks 169 heads-up and 163 nine-handed.

Placed against the full 169, 0 of these 10 hands make the top twenty and 0 reach the top fifty. The middle of the group is 43o at 35.2% (rank 164). In deck terms the group accounts for 112 of the 1,326 possible two-card holdings, or 8.4% of everything you are dealt, so each individual decision carries more weight simply because the spot comes up rarely. The table below lists every hand with its own numbers.

Put AA on the other side of the table — a hand you will run into often — and the best this group can do is 32s at 17.0%, while the worst is 72o at 11.8%. That is a 5.2-point spread inside one hand type, and 0 of them still hold even money or better against it. Grouping hands is useful shorthand, but once you know the specific opponent, the individual figure is the one that matters.

The biggest step in the ordering falls between 32o at 32.3% and 32o at 32.3%, a drop of 0.0 points, and that is where the group effectively splits into an upper and a lower tier. Elsewhere neighbouring hands are often separated by less than half a point, so remembering where the real step is will serve you better than memorising exact positions in the ranking.

Everything on this page assumes the opponent holds a random hand and that the pot is played to showdown. Real opponents hold ranges and can be made to fold, so read these figures as a measure of raw strength rather than as advice on how to play the group. Each hand was simulated over 400,000 showdowns against a single opponent, and every hand-versus-hand figure enumerates all suit layouts before running 200,000 simulated boards. The method page sets out how those numbers were checked against exact enumeration.