There is one decision in Rummy that decides more hands than any other, and it is not which card to take or throw. It is the moment you stop.
Knock too early and you win a trivial amount, or worse, get undercut and pay for the privilege. Knock too late and someone else goes out while you are holding a hand full of unmatched cards. The gap between those two errors is narrower than it looks, and finding it is mostly arithmetic.
What deadwood actually costs
Deadwood is every card in your hand that is not part of a meld, counted at face value with face cards at ten. In the knock-based variants you may end the hand once your deadwood falls to ten or less, and you score the difference between your deadwood and your opponent's.
So a knock with eight deadwood against an opponent holding twenty-two scores you fourteen. Modest. The same knock against an opponent who has melded almost everything and holds four scores you four - and if they hold three, they undercut you and score the difference plus a bonus instead.
That asymmetry is the whole decision. A knock is not a reward for having a tidy hand. It is a bet that your opponent's hand is messier than yours.
The three questions before knocking
1. How many turns has this hand run? Early knocks are safe because nobody has had time to organise. A knock on turn four with ten deadwood is almost always right; the same knock on turn fifteen is a coin flip, because fifteen turns is enough for anyone to have built something.
2. What has the opponent been taking? Someone drawing exclusively from the stock is probably struggling. Someone who has taken three cards from the discard pile has three melds forming and you should knock now if you can, or not at all.
3. Can you realistically improve? Deadwood of eight that consists of one card is worth holding for a draw or two - one useful card takes you to zero and a much larger score. Deadwood of eight spread across three cards will not improve quickly, so take the small win.
Gin, and why it is usually not worth waiting for
Going gin - melding every card, no deadwood - pays a bonus on top of the difference. It is the aspirational play and it is overvalued by most players.
The reason is exposure. Every extra turn you spend converting a good hand into a perfect one is a turn in which your opponent can go out first, and the penalty for being caught mid-conversion is the entire hand. The bonus for gin is fixed; the risk of waiting grows with every turn.
The practical guidance: chase gin when you are one card away and it is early, and take the knock otherwise. Players who consistently chase gin from two cards away lose more to being beaten to the punch than they ever gain in bonuses.
This is a straightforward risk-versus-reward judgement of the kind that shows up wherever a game lets you bank a modest certain gain or push for a larger uncertain one. It is the same decision as whether to take the safe scoring box when you play Yahtzee online - the upper-section bonus tempts you into chasing, and the players who quietly bank reliable scores tend to finish ahead of the ones who chase.
Defensive holding
If you have concluded you cannot win the hand, your goal changes: minimise deadwood so that when someone else knocks, you pay as little as possible.
That means dumping high cards even when they are part of a promising combination. A king-queen of the same suit is a lovely thing to hold and it is twenty points if the hand ends now. Break it up.
Recognising a lost position early and switching to damage limitation is a skill in itself, and it is badly underused. Most players play every hand as though winning it is still possible, which turns a five-point loss into a twenty-five-point one.
The same instinct governs the moment a partnership game turns sour: when you play Spades online and your bid is clearly unmakeable, the correct play stops being "make the bid" and becomes "concede it cheaply and avoid the bags". Accepting the small loss is a learned behaviour.
Reading the moment to strike
Put the pieces together and a pattern emerges. The strongest knocks happen in a specific window: after you have formed two melds, before the opponent has taken more than one or two cards from the pile, with deadwood concentrated in a single card rather than spread.
Outside that window you are usually either premature or late. Watching for it means watching the opponent rather than your own hand, which is the general lesson of every two-player card game: your position is only meaningful relative to theirs.
If you want that principle in its purest form, play something with no hidden information at all - over on Checkers, every piece is visible and the entire game is reading what your opponent can do next. Rummy asks the same question with most of the board concealed.
Across the variants
Oklahoma Gin makes the knock threshold depend on the value of the first upcard, so some hands allow a knock only at very low deadwood - it forces the discipline this article is about. Straight Gin removes knocking entirely: you must go gin, which sounds harder and actually simplifies the decision to nothing. Playing both back to back is the clearest way to feel how much work the knock decision was doing.
In the pile-taking variants such as Rummy 500 the calculus shifts again, because a large discard pile is a resource and a player behind on points can recover in one turn. Holding for gin there is far more defensible.
Take it to the board
For one session, knock the moment you are legally able, regardless of how much better the hand might get. Most players find their score improves, which is an uncomfortable but useful thing to learn about their own patience.
The rules hub has the knock thresholds for each variant, the gin rummy explainer covers the basics, and the glossary defines deadwood, undercut and the rest. The main game is where to practise.
For what to throw while you are waiting for that window, see the companion piece on the discard pile.