What Should You Discard in Cribbage?

The best cribbage discard keeps the four cards that score most on average after the cut, adjusted for whose crib it is. Add what your two throws are worth when the crib is yours, and subtract it when it is the dealer's. Feed your own crib, balk the opponent's, and remember that board position can override the number.

What is the expected average, and why is it the number that matters?

The expected average is Michael Schell's term for a discard's real worth. As dealer, add the crib's average to the hand's; as pone, the non-dealer, subtract the opponent's crib average. The average hand is the kept four's value averaged over every possible starter; the average crib is the crib's expected score given your two throws.

Schell's method scores the kept four against all 46 unseen cards, weighting each rank by how many copies are still out, then reads the crib term from a discard table. The Nobs engine, source of every engine figure here, computes the same quantity exactly and ranks all fifteen ways of splitting six cards. The odds page linked at the end describes how the data set was built; page-specific methods are stated where they are used. The tables are read from that data set, src/_data/odds.json, and the hands were run with the command below.

Schell's example is the dealer's 2♣ 3♣ 5♣ J♣ with an off-suit 4 and 5. It comes out of the engine in his order, and the double run's average hand matches his 12.48 exactly.

Two deals from the odds data set follow. The four-card heart flush against a pair of 5s has the same answer in either seat. Low cards beat ten-cards in either seat too: 2-3-4 wins by 0.87 as pone and 0.59 as dealer over the best keep without it.

Why does whose crib it is change the answer?

The crib belongs to the dealer, which is why the same two cards are an investment on your deal and a gift on the opponent's. In the engine's table a pair of 5s is worth 9.02 when the crib is yours and 9.37 when it is the dealer's. The cheapest throw, 10-K, is worth 3.41 and 3.92.

Every one of the 91 rank pairs is worth more in the opponent's crib than in your own. The gap runs from 0.32 for A-4 to 0.96 for 8-8 and averages 0.61. The unweighted means of the two columns are 4.84 and 5.45.

Rasmussen's explanation, from his own records, is discarding bias. Dealers salt their crib and pones balk it, throwing it cards that combine with nothing, so two cards land in better company when the dealer chose the other two. He calls 8-8 the clearest case, averaging 5.49 over 1,271 throws to his own crib and 8.03 over 182 to an opponent's.

The engine's 8-8 sits at 5.63 and 6.59, the widest gap in its table, in the same direction as Rasmussen's and smaller. The model's notes do not say whether the same mechanism is at work, so treat the parallel as suggestive rather than settled.

What should you throw into your own crib?

Rasmussen tells the dealer to give the crib its two best cards unless the hand pays too high a price for them. Fives and two-card totals of five seed a crib because sixteen of the 52 cards count ten. Checked exhaustively with the engine's scorer, every four-card crib plus starter holding a 5, an A-4 or a 2-3 scores at least two. There is no zero among 4,433,280 combinations for the 5, or among 1,301,800 for either of the others.

The ten best throws to your own crib follow, with the hand points Rasmussen gives up for each; his recorded top ten shares nine of them.

4-5 beats 7-8, 6.97 to 6.76, because it works with the ten-cards and with the pone's usual balking throws. Even so, 7-8 is the throw behind his rare cribs of twenty or more. 5-6 beats 5-K, 7.06 to 6.81, because it is worth two going in and meets the pone's common 7-x and 4-x throws.

Below the 5s, Schell finds that a pair other than 5-5 adds surprisingly little: it is worth two going in and needs specific help. Touching cards, adjacent ranks such as 6-7, grow more ways. A pair of kings adds less to your crib than 6-7 does, 5.04 against 5.56, so from A-3-8-9-Q-Q he throws 8-9. The engine agrees at 8.93 net, with Q-Q thrown third at 7.87.

What should you throw into your opponent's crib?

Balking a crib means throwing the dealer two cards that help neither runs, fifteens nor pairs. Rasmussen's premises for it are in the left column below, with the engine's table beside each wherever it can measure the point.

The ten cheapest gifts are all a king or queen beside a card that neither touches it nor makes fifteen or five with it. The gaps run from three ranks to twelve, and Rasmussen's recorded cheapest gifts share eight of the ten.

Schell's reading of the tables sets the order of danger: 5s and two-card totals of five or fifteen, then mid-card pairs, low pairs and high pairs. In his reading, touching mid-cards such as 6-7 give up more than a low pair, because touching cards grow so many ways. The engine agrees: 6-7 costs 6.32 against 6.10 for A-A. A wider mid-card gap is a different case: 6-8 costs 5.76, below the low pair.

Pairing a mid-card with a high card concedes less than pairing a low card with one, and either beats low plus mid. 8-K cannot combine and costs 4.11, while A-K makes fifteen with a 4 and costs 4.28.

Should you ever give your opponent a 5?

Rarely, and only when the four cards you keep are worth far more without it. Every throw containing a 5 costs at least 6.13 in the opponent's crib, the A-5 row. That is 2.21 more than the cheapest gift and about 0.7 above the average pair in that column. Rasmussen's premise is that a 5 stays in your hand whenever the cards allow. Barlow's is that an opponent can rarely afford to pitch one, which is what makes A-9, 2-8 and 3-7 decent gifts. All thirteen rows follow, in both cribs.

The exception is a hand of twelve or more where the 5 is the odd card out, and two deals from a seeded random sample make the case. Holding 2♦ 5♦ 7♠ 8♣ 8♥ 8♦ as pone, keep 7-8-8-8 and throw 2-5. The hand averages 14.26 (12 to 21), the crib term is 6.20, and the net of 8.06 beats the next option, 5-7-8-8 with 2-8 thrown, by 4.13. Holding A♥ 5♦ 7♠ 7♥ 8♦ 9♥ as pone, keep 7-7-8-9 and throw A-5: 14.35 less 6.13 nets 8.22, a 3.72 margin. Anyone with the site's repository can reproduce either row, or any hand on this page, with the command below. Swap in the six cards, and --own for your own crib.

npm run odds:analyze -- "2D 5D 7S 8C 8H 8D" --opponent

What is Rasmussen's Small Hand Rule?

Rasmussen's Small Hand Rule is for the dealer holding eight points or fewer in the six cards, which he says is more than half of all dealt hands. Judge the discard by the combined value of hand plus crib, so the crib is judged as part of the same hand. Send the strong crib-builders over rather than scraping for a point in hand. The rule is for your own crib only. He adds: plan as if the starter will not help, since most cuts miss, and treat a good cut as a bonus.

Run through the engine, his ten example hands come out his way on eight and within 0.12 of a point on the other two. Here and in the sections below, hands given as ranks only were run with mixed suits, so no four-card flush is in play. Rasmussen's X was read as a king. From 8-7-4-6-Q-5 he throws 7-8 and keeps 4-5-6-Q, which the engine scores at 9.78 in hand (7 to 16) plus 6.76 in the crib. That is a net of 16.54, 2.06 clear of the next throw.

What is Double-Run Hypnosis?

Double-Run Hypnosis is Rasmussen's name for falling in love with a double run when a better keep is on the table. As dealer the cure is usually to send 2-3, 5-5, 7-7 or a 5 with a ten-card to the crib instead. As pone he prefers a nine-point suited hand with an escape card when defending, or what he calls a sleeper when attacking, over an eight-point 2-2-3-4 or 3-4-4-5. He does allow that 3-3-4-5 and 3-4-5-5 are worth keeping.

The engine agrees with nine of his eleven stated answers, and the two exceptions are within a quarter point. On his own puzzle hand, 2-5-6-7-7-K, which he leaves open, it sides with the double run. Keeping 5-6-7-7 and throwing 2-K nets 15.68 against 15.46 for 2-6-7-7 with 5-K thrown. The first hand in the table flips as pone, where the A-A-2-3 double run wins easily, 6.95 against 3.76 for the next option.

The trap is not confined to his examples. Dealt 3♦ 4♣ 4♠ 4♦ 5♦ 7♠ as dealer, the 3-4-4-5 double run ranks third: keeping 4-4-4-7 and throwing 3-5 nets 19.52 against 15.81, a 3.70 margin. Dealt A♠ A♣ 2♥ 3♥ 4♥ 9♥, the heart flush 2-3-4-9 with A-A thrown nets 17.79 against 15.48 for the A-A-2-3 double run.

Which classic discard dilemmas have a clear answer?

Rasmussen's five pair questions compare two throws in isolation, so the engine answers them with the crib term alone. Chambers's hands come with a score and a seat, so the answer is the net average. Rasmussen's own records lean the other way on two of the pair questions, to Q-9 over K-6 for its bust rate and to K-9 over K-10 on average. The engine takes K-6 by 0.08 and K-10 by 0.15.

Chambers asks three questions before any discard: who owns the crib, where each player stands on the board, and how many points position demands from each. Many of his answers are position calls, and the engine ignores position, so where the two disagree the gap is the price of his defensive throw. Of the 21 Chambers answers run, the engine's first choice matches nine outright and eight more sit within a point.

The four large gaps between Chambers and the engine are all hands on the opponent's deal at holes 88 to 95, and the cleanest is 5-5-7-7-8-8 at 93 against 92. The number says keep 7-7-8-8 and throw the 5s. But at 93 the pone will almost certainly not go out this deal, while a 5-5 crib can end the game for a dealer at 92. Chambers throws 8-8, and for once the number is beside the point.

When should you keep cards for pegging instead of points?

When the top two options are close, the pegging quality of the keep can decide at no measurable cost, and the bots in our cribbage app Nobs are built that way. After the exact expected average is computed, the position-aware and skill-aware bots add a small texture bonus to each keep. A kept card of pip value four or less earns a quarter point, as does each pair of touching ranks and each kept pair. Four cards totalling eleven, the magic eleven, earn a full point.

A per-bot weight scales the texture bonus, absent for the seven easiest ladder bots, so at most about 1.2 points of hand average is ever traded for a keep that pegs. The bonus is a taste, not a strategy swap, and never applies to an endgame count-out.

The out card, also called the break card, is the human version of the same idea. It is one card well separated from the other three, the 9 in A-2-3-9 or the Q in 5-6-7-Q. It keeps you out of a forced pegging run.

On the first deal Chambers keeps A-2-9-10 from A-2-9-10-Q-K for its pegging and throws Q-K, which sits 0.10 behind the engine's A-2-10-Q with 9-K thrown. Schell allows that position may dictate A-3-8-9 over A-3-Q-Q as the better pegging hand, at a cost of 1.06. The app's own ruling stands: maximizing points per hand is not always the goal.

Why do ragged hands like 6-7-8-A-A score well?

Raggedy Ann is the American Cribbage Congress glossary's name for 8-7-6-A-A, worth thirteen, and Raggedy Andy is 8-7-6-2-2, worth eleven. Both are five-card totals, the kept four plus the starter; the ACC adds that Ann is also called a ragged thirteen. The hands look like nothing, with no 5, no ten-card, no flush and no jack, yet they outscore a double run. A ragged thirteen is a real score, unlike the nineteen hand, which is slang for nothing at all.

The engine confirms both totals whichever of the five cards is the starter. Mid-cards combine, and 6-7-8 is five points on its own, from 7+8 and the run. An ace or a 2 then fills a fifteen with one of the pairs inside it: 8+6 needs a 1 and 7+6 needs a 2. Each ace sits in two fifteens and the pair, so the second ace lifts 8-7-6-A from seven to thirteen; the second deuce lifts 8-7-6-2 from seven to eleven.

As a bare four-card keep averaged over the 48 other cards, 6-7-8-A scores 9.44 and 6-7-8-2 scores 9.35, against 7.15 for 6-7-8-K. The ragged low card is worth about two points more than a ten-card. Dealt 8-7-6-A-A-K, the engine keeps 8-7-6-A and throws A-K in either seat. That is 9.41 in hand, for a net of 13.18 with your own crib and 5.13 with the opponent's.

How does board position change the discard?

The board-position page covers position in full. Playing on, offence first, means taking the expected-average throw and accepting the crib risk. Playing off, defence first, means spending one or two hand points on the cheapest gift, chosen by the score.

Schell's three lists are the fewest average points for normal play, the throw most likely to bust the crib when a specific count must be stopped, and the fewest big cribs when protecting a lead. Rasmussen's K-10 or K-9 is the model case: K-10 to blank the crib, K-9 to cap it.

Colvert's first-deal postures set each seat's default. The first-hand non-dealer keeps the hand that scores most even if it hands the dealer a good crib. The first-hand dealer plays defence and holds magic-eleven cards against a ten-card lead. The averages behind that split are on the dealer page.

The bots do the same mechanically. In defence a position-aware bot multiplies the opponent-crib term by 1.25. Once either player can realistically finish the deal, the dealer within 29 of 121 or the pone within 21, discarding switches from expected average to the probability of counting out. From there Barlow's rule applies. As pone needing an exact cut to go out, throw the dealer the cards that cut helps; needing a cut only to get close, throw the cards it does not. The endgame page works through it.

Where are the full tables?

The full 91-pair model, both columns, lives in the site's odds data set. The odds page explains how the data set was built, prints the model's extremes beside the crib and hand score distributions, and cross-checks the model against simulated play. The model is the engine's own rather than any published table.

Schell's account of the lineage covers the three tables the engine's model sits beside: Hessel's computer iteration, the table in Colvert's book, and Rasmussen's over-the-board records. In Rasmussen's 87,111 recorded discards to opponents' cribs, a throw as rare as 5-5 came up only 58 times. A discard analyzer for this site is planned; until it ships, the command above reproduces any hand on this page.

By the Nobs Cribbage team Last reviewed 2026-09-03