Gin Rummy Strategy: When to Knock, Tested Over 10,000 Hands

The best gin rummy strategy is to knock early, then switch to playing for gin. When we had Poozee's computer players play 10,000 hands, knocking at the first chance paid off only while 23 or more cards were left in the stock (about your first four turns). After that, knocks were undercut 21% to 80% of the time, and holding out for gin scored more.

The rest of this guide covers what we measured: how often knocks get undercut, when to take a discard, and which cards feed your opponent. Every number comes from the computer players in Poozee's Gin Rummy, so you can check the advice against them.

How we tested

All the tests use the rules Poozee plays: knock with 10 or less, a 25-point bonus for gin and for an undercut, layoffs after a knock but not after gin, and a draw when only two cards are left in the stock.

Unless we say otherwise, both players are our Hard computer. It melds its hand in the best possible way (the game works out the least deadwood exactly), weighs which cards could still make melds, takes a discard only when it makes a meld, and avoids discarding cards that fit what its opponent picked up. To test a habit, we changed just that one habit and played the same deals from both seats, so the luck of the cards cancels out. Scores are points per hand: what you won minus what the other player won.

To test a single decision, like knocking right now, we played the same position both ways and compared the results. A computer isn't a person, but the patterns are large and consistent. A number marked with * is within the normal ups and downs of the sample (less than two standard errors from zero), so treat it as a tie.

When to knock: early, then hold out for gin

We pitted four knocking rules against each other on 4,000 deals, each played from both seats:

• Knock ASAP: knock as soon as your deadwood is 10 or less.
• 5 or less: knock only with 5 or less deadwood.
• Gin only: never knock; hold out for gin.
• Early, then gin: knock at the first chance while 23 or more cards are left in the stock, then only go gin.

The table shows the row player's average points per hand against the column player:

Early, then gin beat every other rule, including the common advice to knock as soon as you can.

Why does waiting work so well? The gin and undercut bonuses are 25 points each. When both players knock as soon as they can, a hand is worth 16.9 points on average, less than either bonus. A player who keeps improving their hand past the knocking point often ends up with so little deadwood that they undercut the knock, or go gin.

Here's the same question as one decision. In 10,000 hands, every time a player could knock, we played the hand out twice: once knocking right away, once playing on (with both players then following Hard's rule). This is the average gain from knocking now, by your deadwood and by how many cards were left in the stock:

Early on, knock with anything up to 10: your opponent has had only a few draws to improve. From the middle of the hand on, knocking lost points even with 1 to 3 deadwood.

Pro tip: The stock starts with 31 cards. Poozee shows how many are left under the stock, so the 23-card mark is easy to watch.

How often a knock gets undercut

An undercut happens when the defender's deadwood, after laying off, is equal to or lower than the knocker's. From the same knocks, here's how often each deadwood count was undercut:

These are hands between two strong players who both hold out for gin once the early turns are over, so late in the hand the defender usually has very little deadwood. Here's how it happens. You knock with 7 in the middle of the hand, and your opponent lays off two clubs on your run:

Undercut: 7 - 6 + 25 = 26 points to your opponent.

Now swap your 7 of spades for the 10 of diamonds. That's gin, and there are no layoffs on gin, so the same two clubs stay in your opponent's deadwood:

Gin: 25 + 19 = 44 points to you.

That's a swing from 26 points for your opponent to 44 for you. Laying off matters for you as a defender too: always check the knocker's runs for cards that extend them at either end.

Count your deadwood right

A card can only be in one meld, so some hands can be melded more than one way. The best way isn't always the obvious one:

Poozee always shows your hand melded the best way, with your deadwood below it. When you're playing with real cards, count both ways whenever a card fits a set and a run.

Take the discard only when it makes a meld

In 5,000 hands, every time the top discard would lower a player's deadwood, we played the hand both ways, taking it or drawing from the stock (or passing, on the first upcard):

Taking a card that makes a meld is a big gain. Taking a low card just to replace a high one cost points later in the hand: you give up a random draw that could have made a meld, and you show your opponent what you're keeping. As a whole habit (on 4,000 deals from both seats), taking every discard that lowers your deadwood cost 0.98 points per hand compared with Hard's rule, and never taking a discard (except to knock at once) cost 8.18.

Which discards feed your opponent

We counted how often each discard was picked up by the other player on the next turn, in 10,000 hands where the discarder didn't try to avoid feeding them. Overall, 19% of discards were taken. By rank:

Low cards get picked up far more, most likely because players throw their high cards first, so the hands they're building are full of low cards. What you've seen your opponent do matters even more:

1. A card that matches one they picked up (the same rank, or within two in the same suit) was taken 38% of the time. Hold it if you can.
2. A card with no connection to what they've shown was taken 27% of the time.
3. A card that matches one they threw away or passed on (the same rank, or next to it in the same suit) was taken 11% of the time. These are your safest discards.

Hard's habit of avoiding cards that fit its opponent's pickups was worth 0.47 points per hand, measured on 20,000 deals (the effect is small, so it needed a bigger sample).

Keep cards that can still make melds

When two discards leave the same deadwood, keep the one with more ways to grow. A pair of 6s or the 8 and 9 of spades can each turn into a meld with one card, while a lone king needs two. Hard weighs this by counting, for every card it hasn't seen, how much drawing it would help. Switching that off, so it simply kept the least deadwood, cost 1.47 points per hand (4,000 deals from both seats).

How long a hand lasts

Between two Hard players, a hand lasted 16.5 draws on average, counting both players and both piles. Hands were split: 42% ended within 10 draws, usually with an early knock, and 42% went 20 draws or more. Of 10,000 hands, 48% ended in gin, 43% in a knock, 1% in an undercut and 8% in a draw.

Between two players who knock as soon as they can, hands were shorter: 11.0 draws on average. 94% ended in a knock, 4% in an undercut, 2% in gin, and only 5 hands in 10,000 ran the stock down to a draw.

Practice against the computer

On Poozee's Gin Rummy, Hard plays the way this guide recommends, and Hint shows what it would do: take or leave the discard, which card to throw, when to knock, and what to lay off. Medium knocks as soon as it can, and Hard beat it by 2.86 points per hand. Medium beat Easy by 6.66.

Keep playing

• Play Gin Rummy against the computer
• Hearts strategy and Spades strategy, tested the same way
• Euchre strategy, for when you have four players
• Play Solitaire on your own