Craps Rules

Key Takeaways

• The whole game hangs on one loop: come-out roll (7/11 wins, 2/3/12 loses, else a point is set) → shooter re-rolls until point (pass wins) or 7 (pass loses).
• Two dice give 36 combinations; 7 is the most common total (6/36) — the game's entire architecture exploits that asymmetry.
• Pass line edge: 1.41%; don't pass: 1.36%. The free odds bet behind either is the casino's only zero-edge wager, cutting combined cost to ~0.37% at 3-4-5× odds.
• The center of the layout is a museum of bad prices: hardways 9–11%, any 7 16.67% — the best and worst bets in the house share one table.
• All craps "strategy" reduces to bet selection; dice have no memory and setters/rhythm rollers have never demonstrated a measurable effect under controlled conditions.

How Craps Works

The dice arithmetic first

Every craps number is a sum of two dice. Combinations out of 36:

Seven sits at the peak with 6/36 = 16.67%. Memorize the triangle and every number in this article becomes checkable by hand.

The pass-line loop

1. Come-out roll. Pass line bets win even money on 7 or 11 (8/36 = 22.2%), lose on 2, 3 or 12 (4/36 = 11.1%), and otherwise — 24/36 = 66.7% of the time — the total (4, 5, 6, 8, 9 or 10) becomes the point.
2. Point phase. The shooter rolls until either the point repeats (pass wins) or a 7 appears (seven out — pass loses, dice pass to the next shooter). Nothing else matters to the pass line.

Don't pass is the mirror: it loses on come-out 7/11, wins on 2 or 3 (12 is "barred" — a push, which is exactly where the casino keeps its edge on the dark side), and during the point phase wins on the 7.

Come / don't come bets are the identical loop started mid-hand: place one after a point exists and the next roll is its private come-out. Mechanically nothing new — just the pass-line game time-shifted.

Free odds: the zero-edge exception

Once a point is set, you may back your pass bet with odds — an additional wager paying the true probability of the point beating the 7:

Payout equals probability: house edge exactly 0%. Casinos cap odds as a multiple of the flat bet — "3-4-5×" (3× on 4/10, 4× on 5/9, 5× on 6/8) is the common structure. The flat bet's edge is fixed, so odds dilute the blended cost:

Pass alone:               1.41%
Pass + 3-4-5× odds:       0.374% of total money wagered
Don't pass + laid odds:   0.341%

No other mainstream casino wager gets a recreational player under half a percent. This is the entire "craps strategy" canon in three lines.

The pass-line edge, derived honestly

Pass wins: 22.2% immediately, plus each point's chance of being set then made — e.g. point 4: (3/36) × (3/9) = 2.78%. Summing all paths gives P(pass wins) = 244/495 = 49.29%, hence:

EV = 0.4929 − 0.5071 = −0.0141  → 1.41% house edge

The don't-pass computation with the barred 12 lands at 1.36%. Both results are exact — craps is fully solvable from the 36-combination table.

The rest of the layout (and what it costs)

• Place bets — back a number directly without the come-out: 6/8 pay 7:6 (1.52% edge — respectable), 5/9 pay 7:5 (4.0%), 4/10 pay 9:5 (6.67%).
• Field — one-roll bet on 2,3,4,9,10,11,12: 5.56% edge with 2:1 on both 2 and 12; 2.78% where the 12 (or 2) pays 3:1.
• Big 6 / Big 8 — the 6/8 place bet paid at even money instead of 7:6: 9.09%. Same event, worse price; a standing lesson in reading the felt.
• Hardways — a pair (e.g. 3+3) before any 7 or "easy" combination: hard 6/8 9.09%, hard 4/10 11.11%.
• Propositions (one-roll) — any craps 11.1%; aces or twelve 30:1 vs true 35:1 = 13.9%; any 7 pays 4:1 vs true 5:1 = 16.67%, the worst standard bet on the table.

The pattern is uniform: excitement density and house edge rise together toward the table's center, while the cheap bets hug the rail.

Variants and the modern table

• Crapless craps — the come-out cannot lose: 2, 3, 11 and 12 become points instead of instant results. Sounds player-friendly. The math disagrees. Trading the 11's instant win (2/36) for four terrible points (2 and 12 are 6:1 underdogs against the seven) raises the pass-line edge to 5.38% — nearly four times the standard game. The variant is a standing exhibit of why "softer rules" require reading the whole ledger.
• Bubble craps — automated dice in a dome with electronic betting terminals: standard math, lower minimums, no crew, and a gentler on-ramp for players intimidated by the live table.
• Online RNG craps — the full layout driven by certified RNG; typically single-player, which removes both the crowd and the shooter ritual. Odds structures mirror land rules, though odds caps vary by title.
• Provably fair dice — crypto casinos' minimalist cousin (roll under/over a target at adjustable odds); not craps by rules, but the same two-dice or uniform math with seed-verifiable rolls.
• Fire bet and modern side bets — longshot propositions on a shooter making 4–6 distinct points pay 249:1/999:1 with house edges in the 20%+ region; small-stakes lottery garnish on an otherwise cheap game.

A worked session

One hour at a moderately busy table (~100 rolls, roughly 30 pass-line decisions), $10 flat bets:

The first two rows are the point worth memorizing: taking full odds quadruples the action while adding barely 25 cents of expected cost — entertainment throughput at near-zero marginal price. The last two rows show how a single "small" recurring center bet, not the pass line, quietly becomes the session's real expense.

Craps vs the other table games

Craps with full odds (~0.34–0.37%) undercuts blackjack basic strategy (~0.5%), baccarat Banker (1.06%) and French roulette (1.35%) — at the cost of the noisiest, most jargon-dense table in the building. The barrier to craps is social and terminological, not mathematical: the optimal game is two bets and no decisions. Online and RNG craps collapse that barrier (no crowd, guided interfaces), and provably-fair dice implementations in crypto casinos are conceptually the same 36-combination game with verifiable rolls.

For Players

• The complete beginner script: pass line at minimum → take maximum odds on the point → repeat. Optionally mirror it with don't pass. You are now playing better craps than most of the table.
• Odds first, always. Any bankroll destined for the layout returns more through odds than through any other bet; size the flat bet small and the odds large within the cap.
• Treat the center as entertainment pricing. A $5 any-seven costs 83 cents in expectation per roll; know it, and either pay it knowingly for the shout or skip it.
• Streaks are theater. A "hot shooter" is survivorship; the dice reset to the same triangle every roll. Dice-setting claims have never survived controlled measurement — bet as if they can't, because they don't.
• Etiquette basics get you comfortable fast: cash on the felt (not into hands), bets down before "no more bets", dice thrown to hit the back wall, keep hands up when the dice are out.

For the Industry

• Craps monetizes complexity. The blended hold on a real table far exceeds 1.41% because live play distributes action across place, field and proposition bets; the felt is a merchandising layout for edge tiers. Odds caps (3-4-5× vs 10×/100×) are the operator's real pricing lever on sophisticated players.
• The onboarding problem is the growth problem. Craps skews old and intimidates novices; guided RNG versions, bubble craps and stream-friendly live formats all attack the same funnel constraint. Interfaces that default to pass+odds and hide propositions behind a tab measurably improve novice retention economics.
• RNG craps needs presentation honesty. Simulated dice must map to a certified uniform 36-outcome model; "rhythm" or physics theatrics in the animation layer must not imply controllability. In provably-fair implementations, publishing per-roll seeds is the trust equivalent of the back-wall rule.
• Catalog note. The SlotCodex database tracks dice games (344 titles as of July 2026) spanning classic craps, sic-bo-style and provably-fair dice; craps proper is a small, high-recognition subset best modeled with rule-variant flags (odds caps, crapless variants) rather than as a monolith.