Not sure which to choose? See what we recommend for each type of player.
Crypto roulette RTP and house edge: the 36/37 rule
A single-zero wheel has 37 pockets and pays as if it had 36. A bet covering n pockets pays 36/n − 1 to 1, so a winning unit comes back as 36/n units, and it wins n times in 37. Multiply the two and every bet on the layout returns 36/37 of what is staked: 97.30%, a house edge of 2.70%. Add a second zero and the same payouts are spread over 38 pockets, which returns 36/38, or 94.74%.
| Bet | Pockets | Pays | Wins, one zero | Edge, one zero | Wins, two zeros | Edge, two zeros | Swing per unit, one zero |
|---|---|---|---|---|---|---|---|
| Straight up | 1 | 35:1 | 2.70% | 2.70% | 2.63% | 5.26% | 5.84 |
| Split | 2 | 17:1 | 5.41% | 2.70% | 5.26% | 5.26% | 4.07 |
| Street | 3 | 11:1 | 8.11% | 2.70% | 7.89% | 5.26% | 3.28 |
| Corner | 4 | 8:1 | 10.81% | 2.70% | 10.53% | 5.26% | 2.79 |
| Six line | 6 | 5:1 | 16.22% | 2.70% | 15.79% | 5.26% | 2.21 |
| Dozen or column | 12 | 2:1 | 32.43% | 2.70% | 31.58% | 5.26% | 1.40 |
| Red/black, odd/even, high/low | 18 | 1:1 | 48.65% | 2.70% | 47.37% | 5.26% | 1.00 |
Arithmetic for 37- and 38-pocket wheels with the standard payout schedule. The last column is the standard deviation of one unit staked: the typical size of a single result around its average.
The edge column never moves; the last column does. A unit on red swings by about one unit either way, a unit on a single number by nearly six. That is the only real choice the layout offers: how rough the ride is. One bet is worse than the rest. On a double-zero table the five-number top line (0, 00, 1, 2, 3) pays 6:1 and costs 7.89%. Mixing bets changes nothing either: chips on several positions are separate bets that each lose 1/37 of their stake on average.
The roulette games reviewed casinos document
"Roulette" on a crypto casino can mean a 37-pocket table, a colour wheel with its own arithmetic, or a studio stream. The in-house games that the casinos reviewed here describe in their own documents:
| Casino | Game | Stated return | Return by arithmetic | Published limits |
|---|---|---|---|---|
| Stake | 37 pockets, 0 to 36 | 97.30% RTP | 36/37 = 97.30% | Minimum $0.0001; profit cap per coin, e.g. 2,500,000 USDT |
| Shuffle | "European Roulette", numbers 0 to 36 | RTP 97.30%, house edge 2.70% | 36/37 = 97.30% | $30,000 bet, $300,000 win |
| Cloudbet | European (37) and American (38) | 97.30% European; "97% RTP" American | 97.30%; 94.74% on a 38-pocket wheel at the payouts listed | Not in the explainer |
| Winna | European wheel, 37 pockets | Published RTP 97.3% | 36/37 = 97.30% | $25,000 across all bets, $360,000 win |
| Thrill | One number from 0 to 36 | None stated | 36/37 at standard payouts | $150,000 in total, $4,000 on a single number; $840,000 payout |
| Rainbet | Roulette Original | "Max RTP 97.30%" | Not computed here | Shown in the game |
| BetFury | Roulette original | 97.30% on the game tile | Not computed here | $7,000 bet, $100,000 payout on originals |
| BitStarz | Roulette Original | 97.3% RTP | Not computed here | Maximum win $108,000 |
| BC.Game | "single-zero table" | 97.3 in the game data | Not computed here | Not stated |
| Wild.io | Roulette Original, 36x top payout | 97.7% on the Originals page; about 97.3% in its Academy | 36/37 = 97.30% | $0.15 to $900 a bet; $45,000 win |
| Roobet | 15-tile colour wheel | None stated | 14/15 = 93.33% on every bet | $100,000 bet, $3,000,000 payout |
| Gamdom | 101-number colour wheel | 99% after a daily allowance | 100/101 = 99.01% on red or black | $600,000 on a colour, $200,000 on green |
| 500 Casino | Colour roulette | About 6.66% edge on every colour | Not computed here | $100,000 single bet; $14,285 on green |
| Rollbit | X-Roulette | 5% house edge | Not computed here | Not stated |
Stake: game page and fairness documentation (1 Nov 2026), help centre (13 Nov 2026). Shuffle: game page and help centre (13 Nov 2026). Cloudbet: provably fair Roulette explainer (13 Nov 2026). BC.Game: game data (13 Nov 2026). Winna: rules page (1 Nov 2026) and help centre (11 Nov 2026). Roobet and Gamdom: see their pages, linked in the first column. Others: as documented in our reviews (11 Nov 2026).
Two entries do not add up on their own figures. Cloudbet's explainer gives American Roulette a "97% RTP" and lists the standard payouts beside it, 35:1 on a single number among them. On 38 pockets those payouts return 36/38, which is 94.74%; a 97% return would need a single number to pay about 35.86 to 1. Its European figure of 97.30% is exact. Wild.io prints 97.7% for a wheel whose top payout is 36x, where 36/37 is 97.30%, the figure its own Academy lesson gives. In both cases the paytable in the game decides, and the arithmetic takes a minute: multiply what a single number returns by the chance of hitting it.
The colour wheels are different games that share the name. Roobet's pays 2x on seven tiles of fifteen and costs 6.67% on every bet; Gamdom's pays 2x on fifty numbers of 101 and costs 0.99%. A strategy article written for a 37-pocket table does not carry over to either, and the gap between them is larger than the gap between European and American roulette.
La partage and en prison
Under la partage, an even-money bet gets half its stake back when zero lands; under en prison it stays on the table for one more spin. Either rule halves the cost of even-money bets on a single-zero wheel: the bet wins on 18 pockets, loses on 18 and loses half on one, which works out to an edge of 1/74, or 1.35%. None of the fairness documents, game pages or help articles cited in the table above mentions either rule for a house roulette, and Stake's stated 97.30% leaves no room for one. Some live-dealer tables from studios do offer it; the rule is printed in the table's own help panel, and it applies to even-money bets only.
Martingale: the arithmetic of doubling
The Martingale bets one unit on an even-money chance and doubles after every loss, so that the first win recovers everything and adds one unit. It wins most cycles. The cost is in the cycle it loses, when the run of losses outlasts either the bankroll or the table limit. On a single-zero wheel an even-money bet loses 19 times in 37, so the chance of losing n in a row is (19/37)n.
| Bets the bankroll or limit allows | Last bet (units) | Lost if every bet loses (units) | Chance a cycle ends that way | Average loss per cycle (units) | Average staked per cycle (units) | Cycles before a bust is more likely than not |
|---|---|---|---|---|---|---|
| 5 | 16 | 31 | 1 in 28 | 0.143 | 5.28 | 19 |
| 7 | 64 | 127 | 1 in 106 | 0.205 | 7.59 | 73 |
| 8 | 128 | 255 | 1 in 207 | 0.238 | 8.80 | 143 |
| 10 | 512 | 1,023 | 1 in 784 | 0.306 | 11.31 | 543 |
| 12 | 2,048 | 4,095 | 1 in 2,974 | 0.377 | 13.95 | 2,061 |
| 15 | 16,384 | 32,767 | 1 in 21,964 | 0.492 | 18.20 | 15,224 |
Arithmetic for an even-money bet on a 37-pocket wheel, starting at one unit. A cycle ends at the first win or when the allowed bets run out.
Read the fifth and sixth columns together. With ten bets available a cycle stakes 11.31 units on average and loses 0.306, and 0.306 ÷ 11.31 is 2.70%. The same division gives 2.70% on every row. The progression has not reduced the house edge at all: it has changed the pattern of results into many one-unit wins and an occasional loss of hundreds or thousands of units, while the average cost per unit staked stays exactly 1/37.
The common rule of thumb that seven or eight losses in a row is a 1 in 128 or 1 in 256 event comes from a coin flip. On a wheel with a zero it is 1 in 106 for seven and 1 in 207 for eight, a fifth to a quarter more likely. Ten in a row is 0.1275%, or 1 in 784.
Where the published limits stop a doubling run
A progression ends when the next bet is not allowed. The published limits of the in-house tables turn into a number of doublings:
| Casino | Limit used | Starting bet | Bets before the next double is refused | Chance of that many losses in a row | Lost by then |
|---|---|---|---|---|---|
| Shuffle | $30,000 maximum bet | $1 | 15 | 1 in 21,964 | $32,767 |
| Shuffle | $30,000 maximum bet | $10 | 12 | 1 in 2,974 | $40,950 |
| Winna | $25,000 across all bets | $1 | 15 | 1 in 21,964 | $32,767 |
| Wild.io | $900 maximum bet | $0.15 | 13 | 1 in 5,792 | $1,228.65 |
| Thrill | $150,000 in total | $1 | 18 | 1 in 162,206 | $262,143 |
Arithmetic from each casino's published maximum (sources as in the table above). "Lost by then" is the sum of the doubled bets, 2n − 1 times the starting bet.
A $1 Martingale on Shuffle's table can absorb fourteen losses and place a fifteenth bet of $16,384; a sixteenth, $32,768, is over the $30,000 maximum. Reaching that point is a 1 in 21,964 event per cycle, and it costs $32,767 to have collected $1 at a time until then. Start at $10 and the wall is twelve bets away, a 1 in 2,974 event. On Stake the minimum bet is $0.0001 and the ceiling is a profit cap per coin, so the number of doublings is limited by the player's balance long before the table: the twentieth bet of a run that starts at one cent is already $5,242.88.
Fibonacci, D'Alembert and Labouchere on the same ten losses
The gentler systems raise the stake more slowly after a loss. Put each of them through the same ten straight losses, an event with a chance of 0.1275%, and compare what is on the table:
| System | Rule after a loss | Tenth bet (units) | Lost after ten losses (units) |
|---|---|---|---|
| Flat betting | Same bet | 1 | 10 |
| D'Alembert | Add one unit | 10 | 55 |
| Fibonacci | Next number in 1, 1, 2, 3, 5, 8… | 55 | 143 |
| Martingale | Double | 512 | 1,023 |
Arithmetic. Labouchere depends on the list the player writes down: with the list 1-2-3 the bets run 4, 5, 6 and so on, and ten losses cost 85 units.
The slower a system raises its bets, the less a bad run costs and the less a single win repairs. D'Alembert is 55 units down after ten losses and wins back eleven on the next spin; Fibonacci is 143 down and wins back 89, the size of its eleventh bet. Neither result says anything about the average, which is the same for all of them: every chip any system places loses 1/37 of its value on average, so the expected loss is 2.70% of the total staked, whatever order the bets come in. A system that stakes more in total costs more in total.
The reason no ordering can help is independence. On the in-house games each spin is computed from a fresh number, and the previous result is not an input. A pocket that has not appeared for 100 spins has the same 1 in 37 chance on the next one; the worked streak tables are on our Stake Roulette page.
Is crypto roulette provably fair?
The in-house wheels are, in the sense that a spin can be recomputed from values fixed before the bet. How the number is turned into a pocket is documented differently from casino to casino:
- Stake takes one number between 0 and 1 from an HMAC-SHA256 of the server seed, client seed and nonce, and the pocket is that number times 37, rounded down. Its documentation says the game is "derived from the European version of the game where the wheel consists of 37 possible different pockets, ranging from 0 to 36" (1 Nov 2026).
- Shuffle documents the same seed-pair construction for its Roulette in its game-events article, and its game page calls the wheel "European single-zero roulette" at "the classic 2.70% house edge" (13 Nov 2026).
- Cloudbet draws whole numbers from a SHAKE256 stream and applies rejection sampling before taking the remainder: "in European Roulette, one whole number between 0 and 36 is sampled from the random number sequence" (13 Nov 2026). Its explainer prints a Node.js program for checking a spin.
- Thrill also uses rejection sampling for whole numbers, which avoids the tiny bias a plain remainder would introduce.
- Roobet and Gamdom run their colour wheels as shared rounds on a published hash chain, with no player seed; each has its own verifier.
Our roulette verifier reproduces Stake and Shuffle spins, and the general provably fair verifier handles the commitment check. Live-dealer roulette is a different matter: a physical wheel has no seed, and its fairness rests on the studio's licence and equipment testing. If recomputing a spin matters to you, play the in-house game.
Verification settles whether the pocket was fixed in advance. It does not move the 2.70%. The lowest house edge comparison shows where roulette sits against the 1% Originals, and the roulette calculator prices any combination of bets on 37 or 38 pockets.
A flat-stake session, priced
If no system changes the cost, the cost is easy to state in advance. Take 100 even-money bets of $10 on a single-zero wheel: $1,000 staked, an expected loss of $27.03, and a typical swing around that of about $100 either way. The chance of finishing ahead after 100 such bets is 35.5%; after 1,000 bets it is 18.8%, and after 10,000 it is 0.3% (exact binomial arithmetic, with a level finish counted as not ahead). The longer the session, the more the result looks like the edge and the less like luck.
- Pick the wheel before the bet. One zero costs 2.70% and two cost 5.26%. Among in-house games, check which game the name refers to: the colour wheels above range from 0.99% to 6.67%.
- Decide the total you will stake, not the number of wins you need. Expected cost is 2.70% of the total, so the total is the figure to budget.
- Use the casino's own limit tools. A loss limit set in the account holds when a progression would not. Our responsible gambling page lists what to set.
Related: Stake Roulette · Roobet Roulette · Gamdom Roulette · best crypto casinos · provably fair casinos · dice strategy guide · dice verifier
Key takeaways
- Single-zero roulette returns 36/37: 97.30%. Double-zero returns 94.74% at the same payouts.
- Every bet on the layout has the same edge. Bets differ in swing, from about 1 unit on red to about 5.8 on a single number.
- Martingale loses exactly 2.70% of what it stakes. Ten losses in a row is 1 in 784, and it costs 1,023 units.
- Published limits end a doubling run: fifteen bets from $1 at a $30,000 maximum.
- In-house wheels can be verified from seeds; live wheels cannot. "Roulette" also names colour games that cost between 0.99% and 6.67%.