How Do You Play the Cribbage Endgame?
The cribbage endgame begins on Fourth Street, holes 91 to 120, where the goal stops being points and becomes reaching the game hole before your opponent. Count what each side can score in its remaining counts, then discard and peg for that number. The pone counts first at the show, and that order decides most close games.
Why does counting first decide the endgame?
Michael Schell's three counts are the dealer's hand and crib on one deal, then that player's hand as pone on the next, with only that deal's pegging between. Your first count is your hand as pone, scored before the dealer's hand and crib under the counting order. He calls that order a significant late-game advantage. The objective follows: get close enough to 121 that your final three counts take you out.
When both hold enough at the show to go out, the pone wins; the dealer's hand and crib never get counted. The dealer's compensations come earlier: his heels on a jack starter, and the near-certain go or last-card point from playing second.
The top bots in our cribbage app Nobs measure the race to 121 in half-deals. Their endgame search switches on when either side could go out this deal: the dealer from 92, the pone from 100. Inside that search the discard is chosen to maximise how many of the 46 unseen starters count the keep out, rather than expected points. The app's ladder help calls this counting out the endgame.
What is a specific count?
A specific count is the pone's discard on the last deal, chosen to maximise the chance of scoring exactly the number needed. Barlow's discarding quiz gives the rule: keep the four cards with the most winning cuts, not the most average points.
| Measure | Meaning |
|---|---|
| Winning cuts | Starters, out of the 46 unseen cards, that give you enough to go out on the first count |
| Winning ranks | Cut ranks, ace to king, that do the same |
| Close cuts and close ranks | Starters, or ranks, that leave you within a few holes of going out, typically three |
| Supercuts | Winning cuts plus close cuts |
| Superranks | Winning ranks plus close ranks |
| Endgame count | Schell's name for the point-count system built on these measures |
In the two hands below, the pone sits at 112 needing 9 and the dealer at 115 needing 6. The Nobs engine scored every keep on all 46 unseen starters; winning cuts reach 9, close cuts land on 6, 7 or 8.
| Hand A, A♦ 2♣ 2♦ 4♥ J♥ Q♥: keep | Toss | Winning cuts | Which cuts | Close cuts | Supercuts | Hand average |
|---|---|---|---|---|---|---|
| A-4-J-Q | 2-2 | 15 of 46, 32.6% | Any A, 4, 10 or K, plus the 5♥ by nobs | 9 | 24 | 6.74 |
| A-2-2-J | 4-Q | 7 of 46, 15.2% | Either 2, any 3, plus the A♥ by nobs | 23 | 30 | 6.74 |
| A-2-2-Q | 4-J | 6 of 46, 13.0% | Either 2, any 3 | 24 | 30 | 6.52 |
Both top keeps average 6.74, yet A-4-J-Q wins on more than twice as many cuts. Supercuts favour A-2-2-J, but close cuts need a next deal, and a dealer needing 6 will almost certainly end this one.
| Hand B, 5♥ 8♥ 8♠ 9♦ 10♥ J♦: keep | Toss | Winning cuts | Which cuts | Close cuts | Supercuts | Hand average |
|---|---|---|---|---|---|---|
| 5-9-10-J | 8-8 | 29 of 46, 63.0% | Any A, 5, 6, 9, 10, J, Q or K, plus the 8♦ by nobs | 17 | 46 | 9.70 |
| 8-8-9-10 | 5-J | 22 of 46, 47.8% | Any 5, 6, 7, 8, 9, 10 or J | 24 | 46 | 10.30 |
| 5-8-8-9 | 10-J | 7 of 46, 15.2% | Any 7 or 10 | 6 | 13 | 5.00 |
In hand B the higher average is the worse keep; with 46 supercuts each, winning cuts alone decide. Both tables cover the first count only: you may peg the shortfall, or the dealer may peg out first. On hand A, ONYX and the Champions, the bots that run the endgame search, would keep A-4-J-Q. The rest of the ladder, and expected value alone, prefer A-2-2-J.
What is Barlow's rule?
The Cribbage Forum glossary names Barlow's rule for the pone's toss on the last deal. When you need a particular cut to go out, throw the dealer the cards that cut helps. When you need a cut merely to get close, throw the cards it does not help.
Barlow reasons that if the cut puts you out, you count first and the crib is never scored, so that toss costs nothing. When even your best cut leaves you short, the game must go on, so the crib should be empty when that cut comes. His illustrations:
| Your hand | You need | Dealer needs | Toss | Why |
|---|---|---|---|---|
| A-2-6-6-6-K | 12 | 16 | 2-K | The 3 that feeds a 2-K crib also puts you out |
| A-2-6-6-6-K | 16 | 16 | A-K | A 3 no longer puts you out, so the 2 it would help stays out of the crib |
| 2-4-5-6-9-10 | 14 | 17 | 2-9 | A 4 puts you out, so a 4 may help the crib |
| 2-4-5-6-9-10 | 18 | 17 | 2-10 | A 4 does not put you out, so keep it from the crib |
Schell's glossary also records the Colvert exception, a lead rule for desperation defense. Lead a low card, unless a middle card, 6 to 9, carries just one loser more than the low card does; then the middle card goes first. A loser is any card the opponent can play to score on yours. Colvert's reason is that a good opponent keeps every small card he is dealt expecting your low lead, so the middle card is the percentage lead.
How should the dealer discard on the last deal?
The Cribbage Forum glossary gives the last deal's dealer two postures. Desperation defense disregards offense entirely to keep the opponent from scoring, or from scoring a specific number, even at the risk of conceding a big score. Prevent defense is playing off in its narrowest form: it accepts small scores to avoid a large one, typically protecting a big lead late.
When your own count will be enough, keep the hand that caps the opponent's pegging, not the biggest hand. Barlow's tip 20 records a loss, set out below in our own table: his keep let a run through. He adds that he looked at the board only after the lead, so check the position before the first card.
| Barlow's spot | He kept | He met | Better keep |
|---|---|---|---|
| Dealer needing 8 against an opponent needing 10, dealt 2-6-9-9-10-K | 6-9-9-10 | 7-8-9-J, which pegged four in a run | 2-6-9-9 or 6-9-9-K, which make the run avoidable |
When nobody's count matters, Chambers's closing case applies. Level at 118 on your own deal, the size of your hand is irrelevant because the pone counts first. You will have to peg out, so keep four different cards and play on.
What is defensive pegging in the endgame?
Colvert's tip 8 opens with a claim to take as his, not ours. In his view the last few pegs decide more games than everything played before them, and a typical game turns on a margin of six points at most. His defensive rules for the endgame, paraphrased:
| When the opponent needs | Colvert's rule |
|---|---|
| Anything | Lead below a 5, from a pair when you hold one |
| Anything | Dump a lone jack at a safe opportunity |
| Anything | Tally the cards that could beat you, then take the percentage play rather than a hunch |
| Four or more pegs | Do not be caught with one small card when the count is above 21; toss it, or play it early, usually second |
| Five or more | Do not keep a 4, 5 or 6 as your last card |
| Seven or more | A 5 held to the end is deadly: it walks into 4-5-6 and 5-5-5 traps |
Barlow adds three refinements. With enough to go out on your first count but an opponent who could peg out, lead the middle card of a sequence early, so what remains is diverse.
Work out the worst case first, then steer around it. Needing 11 against 9 with 5-9-10-J, Barlow led 10, jack, 9, then 5, and was pegged out by a run plus a go. Playing the 5 before the 9 would have left the dealer no two-card holding able to peg six.
When the opponent needs fewer than three, play so he is least likely to peg at all, not so he pegs the least. A play that could give up five holes is no worse than one that gives up two.
What is offensive pegging in the endgame?
Colvert's tip 9 is the mirror image, for the player who needs the pegs. His rules, paraphrased:
| Pegs needed | Colvert's rule |
|---|---|
| Any | Save for last two cards that score together: a pair, two that make fifteen, or failing that two that could make a run |
| Six or more | Prefer a small pair, aces or 2s, and set a trap once the count is beyond 21; or keep 4-5-6 pieces for a run plus a fifteen |
| Ten or more | Hold a low pair, 7s at most, for four of a kind, and never lead from it |
| Any | Never lead a card that removes your own winners: from 8-9-10-J needing 3, lead from an end, not the 9 or 10 |
| Any | Count the winners before leading |
A winner, in the Forum's vocabulary, is any card the opponent could play that sets up a score for the card you hold next. Colvert's own loss shows the last rule: needing 2 with 4-5-10-J, he led the 4, which had about six likely winners, instead of the 10, which had about ten.
Once the endgame search is on, the top bots in Nobs run offense and defense as one calculation. Each playable card is scored by its average chance of winning across several sampled opponent holdings. The engine's code describes that search as covering both races to peg out and holding the opponent short. The post-game grader shifts the same way: mid-game plays are graded on net pegs, endgame plays on win probability lost.
What is a two-on-one?
A two-on-one, in the Cribbage Forum's usage, is the end of the pegging when the player to play next holds more cards than the opponent. The extra cards guarantee the last-card point and can trap the lone card for a fifteen, a pair or a run. The lone card's owner decides only which card to keep for last, or with two against one, which to play.
Schell's rule for a dealer forced into an early go with two low cards, neither making 31, is to play the lower first and save the higher for last. The lower a card's value, the more two-card holdings can trap it for a fifteen.
| Last card | Two-card holdings that trap it | Ruled out unless a 31 was passed up |
|---|---|---|
| Ace | 7-7, 6-8, 5-9 | A 4 with a ten-card |
| 2 | 6-7, 5-8, 4-9 | A 3 with a ten-card |
| 3 | 6-6, 5-7 | 4-8 |
| 4 | None | 5-6 |
Schell's four exceptions:
| Exception | His case |
|---|---|
| Only one possible trap changes the result: save the card whose traps are already decided | At 117 against 113 with 2-3 left, keep the 2 and play the 3 first; the 6-6 trap on the 3 is the one holding the choice decides |
| The go has proved the pone holds nothing low enough to trap either card | The order does not matter |
| Your own count already takes you out | A 31 is worthless, so think defense |
| Board position overrides | At 80 against 84, play the 3 first: the holdings that trap a 3 are the stronger ones, whose two extra points move the pone's position materially, while the ones that trap a 2 reach only 91 |
Barlow's method with two cards against one sorts the opponent's possible last card into three groups: already won, already lost, and decided by your choice. Play the card that gives nothing to the third group; the right play can never hurt, even if it rarely pays. Barlow's caution is to assume a strong opponent holds the largest hand the cards played allow, and to defend accordingly until proven otherwise.
What is the bitter end?
The bitter end is Schell's name for endgame pegging when you must peg to win. His first Bitter End article says save your pair or your touching cards for last, because a competent opponent denies you scoring chances early. The score chooses between them.
| Pegs needed | Keep for last | Why |
|---|---|---|
| Exactly one | Whatever pushes the count highest: play for the first go | Any go or last card wins |
| Two, three or four | The touching cards | A run, or a pair on one of the opponent's cards, is the likeliest route; a low pair such as A-A that can be played back to back on a go is the exception |
| Six or seven | The pair | Tripling the opponent's card is the only route |
The second article adds a counting rule: follow each line one play past its apparent end. A reply that scores for the opponent but lets you go out at once is not a loser.
At 117 apiece with 4-5-6-7, after a 4 and a 2, Schell counted seven losers on the 6, six on the 7, and played the 7. But the 3 that makes fifteen on the 6 hands him a run worth five and the game, so the 6 had three losers and was right.
Stealing last card is the pone's counter: trap the dealer into playing two cards in a row on one go. The parity of the play flips, the pone takes the last-card point, and the dealer faces a two-on-one on the next series.
How do you play from the stinkhole?
The stinkhole is hole 120, one short of game; the term, and the ACC rulebook's different placement of it, are on that page. From 120 any single point wins, in any phase: a go, last card, a pair, a fifteen, a 31 or his heels. The pone at 120 with any points in hand wins at the first count, unless the dealer pegs out first. Restrictions on scoring out of 120 are house rules, covered under pegging out.
A jack starter scores two for the dealer before a card is played, so a dealer needing two or fewer can win on the cut alone. From the dealer's seat at most 4 of the 46 unseen cards are jacks, 8.7%, and fewer if he holds one. In the 25,000-game corpus described below, heels ended 132 games, 0.5%.
Barlow's lead rule for both players a hole or two from home is in the table below. In one two-against-one position, Barlow held 4-5 needing three. A 3 or 6 in the opponent's hand won for him whatever he led, and a ten-card lost whatever he led; only a 4 or 5 turned on his choice. So he led the rank he judged the opponent less likely to hold, and kept the pair for the likelier one.
| Your cards | Barlow's lead | Why |
|---|---|---|
| Any low card | A 2 or 4 before an ace or 3 | A low lead allows no fifteen; the dealer cannot answer a 2 or 4 without leaving you two pegging cards, but a 6 on a 3 or a 7 on an ace cuts you to one |
| No low card | A ten-card, not a 6 to 9 | Here the dealer keeps mid-cards for pegging and pitches ten-cards to the crib |
When should you play for, or against, the skunk?
A skunk is a win by 31 or more, and the stakes decide the question. In ACC play a loss scores 0 game points, a win 2, a skunk 3, and a double skunk also 3, nothing more. Schell reports the same 2/3/0 scheme in Grass Roots play, qualifying rounds and most Internet match play.
For the trailing player the Cribbage Forum gives three skunk positional holes. Reach each before the opponent reaches the matching positional hole, and you should cross the skunk line, between holes 90 and 91, before the opponent goes out.
| Your skunk positional hole | Opponent's matching positional hole |
|---|---|
| 14 | 18 |
| 40 | 44 |
| 66 | 70 |
For the leader, Schell's answer is to forget the skunk when you cannot go out this deal. Aim instead to maximise your chance of finishing with next deal's three counts. With the opponent far back, that means a prevent defense, conceding a point or two to avoid a big crib. The general rule is our inference: a skunk adds one game point and the win is worth two, so chase the skunk only when the win itself is safe.
Should you ever concede?
Barlow's tip 19 is a story, not analysis. Needing 28 against an opponent needing two with first count, he wanted to concede; his grandfather's tale of a win from 26 behind kept him playing. He lost, but held the opponent to zero pegs before a 24 hand ended it.
The rules settle the rest. A game ends only when a player reaches 121, in any phase, with no exact count needed. The ACC rulebook's provisions, paraphrased:
| Where | ACC rule |
|---|---|
| Every event except match play | Pegging out is mandatory, and the opponent must remind a player to peg out (not in muggins events); refusing forfeits the score while play continues |
| Qualifying rounds | A game cannot be conceded at all |
| Match play | A concession is recognised, and counts as the opponent pegging out |
The corpus is 25,000 games of ONYX, the strongest of the three bots in the
browser game, playing itself from a fixed seed. It is bot-versus-bot play, not
a claim about human games, and the winner's final item is clamped at 121. The figures
are computed from the site's odds data set, src/_data/odds.json, by the Nobs engine;
the method is on the odds page.
| Which count ended the game | Games | Share |
|---|---|---|
| A hand count, pone's or dealer's | 15,284 | 61.1% |
| Pegging | 4,978 | 19.9% |
| The crib | 4,606 | 18.4% |
| His heels | 132 | 0.5% |
Four games in five end at the show, where the counting order applies; the hand row is not split by role. Three rows of the dealer's Fourth Street win rate, from the board-position page, answer the concession question:
| Dealer's deficit at the start of a Fourth Street deal | Dealer still won |
|---|---|
| 20 to 24 behind | 1.2% |
| 15 to 19 behind | 6.5% |
| 10 to 14 behind | 17.5% |
Sources
- Michael Schell, Cribbage Forum glossary. Three counts, first count, the specific-count measures, the play modes, the two-on-one, skunk positional holes and game points.
- Michael Schell, Cribbage Forum: The Bitter End (parts 1 and 2), Defending the Two-on-One, Stealing Last Card. The bitter-end rules, the two-on-one defence and its exceptions, and the stolen last card.
- Michael Schell, Cribbage Forum, Ask the Experts on the opponent's crib. The leader's skunk rule and prevent defense.
- Dan Barlow, ACC tips 7 and 13, discarding on the last deal. Barlow's rule and the winning-cuts principle.
- Dan Barlow, ACC tips 6, 12, 14, 15 and 20, endgame pegging. Leads near home, the worst-case rule, the fewer-than-three rule and the last-deal keep.
- Dan Barlow, ACC tips 1, 17 and 19, two cards against one and conceding. The three-group method, the pessimism rule and the conceding anecdote.
- DeLynn Colvert, ACC tips 8 and 9, endgame pegging. The defensive and offensive pegging rules and the Colvert exception.
- John Chambers, ACC tip 16, Fourth Street. The level-at-118 case.
- American Cribbage Congress, tournament rulebook (2025 edition). Pegging out, concession and game points, paraphrased; the ACC web glossary and director manual agree on 2/3/0.
By the Nobs Cribbage team Last reviewed 2026-09-03