Cribbage Strategy: How Do Good Players Win More Often?

Cribbage strategy is three decisions on every deal: which four cards to keep, which card to play next, and whether the board calls for attack or defence. The deal sets the ceiling. By DeLynn Colvert's estimate, play between average players decides roughly ten points of a 121-point game, and those points are the whole contest.

Is cribbage luck or skill?

Both, in a proportion the experts have tried to measure. George Rasmussen's essay on skill names four things no decision touches: the shuffle, the cut for deal, the way the cards fall and the starter cut. Everything after you pick up six cards is choice.

Michael Schell's course of study puts the split at about two-thirds luck. Two games in three against a competent opponent are settled by the cards, and those split evenly, so the contest is the remaining third. DeLynn Colvert measures that third in points: between average players, ten or so a game, and usually fewer.

One piece of the luck can be measured, because the cut decides who deals first. ONYX, the strongest of the three bots in the browser game, played itself for 25,000 fixed-seed games, and the first dealer won 56.4% (95% interval 55.8% to 57.0%). That is bot-versus-bot play, with the winner's final count clamped at 121. The figure is computed from the site's odds data set, src/_data/odds.json, by the Nobs engine, and the method is on the odds page.

What are the three decisions in every deal?

The three are keep, play and posture. The rules leave a player exactly two choices: which two of six cards to throw, and which held card to play at each turn. A go is not a choice, since it may be called only when nothing fits under 31. The third decision is how the other two are made. Rasmussen's essay walks the game in that order, from counting the hand to discarding to pegging to board position, which he calls the basis of all game strategy.

Schell's glossary separates strategy, the long-term plan, from tactics, the devices of the moment, and defines board strategy as using position to balance offence and defence. In his course of study board strategy is the concept that separates experts from the rest.

Chambers and Rasmussen open every discard with the same three questions: who owns the crib, how far you stand from the hole your position demands, and how far the opponent stands from theirs. The table sends each decision to the page that teaches it.

What should you keep and what should you throw?

Keep the four cards with the highest expected average, then bend the number for position. The expected average is Schell's name for the average hand after the cut plus the average crib when the crib is yours, minus it when it is the dealer's. He calls it the single most important number in discarding, and whenever offence is the priority the best throw is normally the one that maximises it.

In Rasmussen's records of his own crib, 5-5 leads at 8.82, followed by 2-3, 5-J, 5-6 and 5-K. He will spend as many as four hand points to send the pair of 5s over, provided the hand still has twelve-point potential.

Into the opponent's crib Rasmussen throws two cards that build nothing: the king first, the queen next, no jack if he can help it, no 5 if the cards allow, and no touching or one-gap pair. Only 10-K and 9-K averaged under four points in his records. The Nobs engine's model agrees at both ends: 5-5 is worth 9.02 in your crib and 9.37 in the opponent's, and 10-K is worth 3.41 and 3.92.

Position overrides all of it. On the first deal Colvert tells the first non-dealer to keep the biggest count and stop worrying about balking. The discard page tests every classic throw against the engine.

What is the best card to lead in cribbage?

Lead a card below a 5, preferably from a pair. That is Colvert's defensive rule, and the arithmetic behind it is short. No reply can make fifteen on a lead of 4 or lower, while a led 5 gives any of the sixteen ten-cards a fifteen. Colvert counts three losers for a 2, 3 or 4 lead, the cards that pair it, against seven for a lone queen.

Many tables teach the shorter form, lead a 4: Bicycle's rules page names the 4 among the safest leads, and Schell leads the 4 from 4-5-5-6, though as a blanket rule it is commonly taught rather than sourced. With no low card, Barlow leads a face card or a 10, because mid-cards are what dealers keep. Schell adds that any ten-card duplicated by the starter or your own throw is the safest lead.

The simulation shows what a 5 costs. Across 224,294 simulated deals with an opening lead, the dealer's immediate reply scored on 21.8% of the 37,373 leads of a 4. It scored on 71.5% of the 1,167 leads of a 5. The pegging page prints all thirteen rows. That table is policy-conditioned: each row describes the positions ONYX chose to lead that rank from, not the value of a random lead.

What is the magic eleven in cribbage?

A magic eleven is two cards totalling eleven, kept because two ten-cards make 20 and the eleven then makes 31 for two. Colvert's tip lists the pairs, 6-5, 7-4, 8-3, 9-2 and an ace with any ten-card. His reasoning is that ten-cards with 5s are among the commonest holdings, and 5s are rarely led, so the ten-cards come out first. When a discard is a close call he keeps the eleven. For desperation defence with no two-card eleven he gives the three-card elevens, 5-3-3, 9-A-A, 8-2-A, 7-2-2 and 6-3-2.

Rasmussen calls magic elevens Elevens from Heaven, and gives the two-card elevens to the dealer and the three-card elevens to the non-dealer. Schell's glossary completes the family. A magic five, A-4 or 2-3, scores fifteen on a ten-card reply when led from, and a magic sixteen scores 31 against a 5 and a ten-card. The bad eleven and bad sixteen are the same totals in the leader's hand, which give the 31 away instead.

The Nobs bots apply the magic eleven at discard time only, as a bonus for a kept four totalling eleven. The pegging page has the parity rule and the full table.

What is the 26 theory in cribbage?

The Twenty-Six Theory is DeLynn Colvert's board strategy, built on his finding that the dealer and non-dealer together average about 26 points a deal, 26.4 unrounded. He rounds down because slowing a game is easier than speeding it up: runs, pairs and fifteens take both players' cooperation. Projected around the board, the first dealer reaches 120 after nine average deals with first count due on deal ten, while the first non-dealer sits at 114, seven short.

Colvert turns that gap into two jobs. The first non-dealer plays offence from the first card to find seven points above average. The first dealer can give away nine points and still count out first, and so plays defence from the first card to slow the game. About one game in ten runs near average all the way, and those are the games the theory is meant to win. Rasmussen claims that the players who follow a board strategy most closely win more at home and on the tournament trail.

The Nobs bots that read the board use the same per-role paces to project who reaches 121 first, in half-deal steps because the pone counts first. They play offence when clearly behind in that race, defence when clearly ahead, and neutral in between. The board-position page prints the projection deal by deal.

Is it better to go first or second in cribbage?

There is no choice, because the cut decides it: the lower card deals first. The deal is worth having. The engine's first dealer won 56.4% of the 25,000 games above. Colvert's projection at equal skill is 56 in 100, with the first non-dealer about five points short at the unrounded 26.4. Chambers calls the first deal and crib a slight advantage, because the dealer gets three counts on the way to hole 17 or 18 against the pone's single count.

Colvert's charts, from samples of a thousand games or more, put nine-deal games at about 42% of all games. The dealer won those 250 to 170 in his sample, usually by counting the crib. The engine corpus has 44.1% of games ending on deal nine and 90.5% in eight to ten deals.

Second is not hopeless. Colvert calls the first non-dealer's scoring the key to the game. One hand of sixteen or more in nine deals recovers the seven points, while the first dealer would need two such hands to count out a deal early. The averages behind the dealer's edge are on the dealer page, and the board-position page sets Colvert's figures beside the engine's.

How do you play the last deal?

On the last deal the objective changes from points to the game hole. In Schell's terms you want to be close enough to 121 that your three counts take you out: hand and crib as dealer, then your hand as pone. The pone counts first, and the rule on who counts first is on the counting order page.

The pone's last deal is usually desperation offence and the dealer's desperation defence. The pone's throw becomes a specific count, chosen to maximise the winning cuts, the starters that give exactly enough to go out. Barlow's rule follows: needing a 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 help.

Colvert claims that more games are won or lost pegging the last few points than by all the earlier play. He also claims that the typical game turns on a margin of six points at most. Schell's bitter-end rule is to save the pair when you need six or seven and the touching cards when you need two to four. When either player could realistically finish this deal, the Nobs bots stop maximising points and maximise the chance of winning. The endgame page works through all of it.

Which named strategies do the pros use?

The names below are the ones used by the five authors of the American Cribbage Congress tip library, Barlow, Chambers, Colvert, Rasmussen and Schell, and by Schell's Cribbage Forum. The ACC library is organised by author rather than topic, so the same idea can carry three names and a single page can mix discarding, pegging and position. The table groups each play by the decision it serves, names who teaches it, and links the section of this site that works it through. One row is table lore with no author behind it, and says so.

How do you get better at cribbage?

Read, keep a database, and spar against something strong. Schell's course of study starts with Colvert's Play Winning Cribbage, the one book if you read only one. Next comes Chambers's Cribbage: A New Concept, for board strategy, then Barlow's Cribbage for Experts and Miracles on Fourth Street.

Schell's second step, borrowed from chess and backgammon, is a personal database of archetypal six-card hands for discarding and four-card hands for pegging. Each new play goes into the database and is tried at the next opportunity. His third step is sparring against a strong program that logs your games, and his targets for an improving player are 55% online and 52% in Grass Roots.

The bots in our cribbage app Nobs are built for that third step, the sparring. A bot sees only its own hand, the played cards and the starter, and every deck is shuffled honestly for both sides. A harder bot plays better rather than getting better cards: smarter discards, careful pegging, memory, board sense and, at the top, pair traps, magic elevens and counting out the endgame.

Every finished game is graded on discard accuracy and pegging accuracy. Beside those sits a card-luck meter, which measures which player the cards favoured and says nothing about how well you played. A discard trainer grades your throw from six cards against all fifteen options.

Where do you go from here?

Four pages carry the depth, in the order the decisions arise in a deal.

• What should you discard in cribbage? The expected average, both crib guides, Rasmussen's rules and the classic dilemmas, each run through the engine.
• What is good cribbage pegging strategy? Leads, counts to avoid, the magic eleven, losers and winners, traps and reading your opponent, with the simulated lead table.
• How does board position decide your cribbage strategy? The 26 theory, par holes, positional holes, the streets and the dealer's win rate from every position.
• How do you play the cribbage endgame? The specific count, Barlow's rule, the two-on-one, the bitter end, the stinkhole and the skunk.

If the rules themselves are still new, start with how to play cribbage and come back. Every page here assumes you can count a hand and know who counts first.

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