Roulette Odds and Payouts: What Every Bet Pays and What It Costs

Why the answer is always the same number

This is the part most payout charts leave out, and it is the only thing on the page worth memorising. Roulette payouts are calculated as though the wheel had 36 pockets. A straight-up number pays 35 to 1, which would be exactly fair on 36 pockets. A corner covering four numbers pays 8 to 1, which is 36 divided by 4, minus 1. A dozen pays 2 to 1, which is 36 divided by 12, minus 1.

The zero is simply not accounted for anywhere in that schedule. It sits on the wheel taking one pocket out of 37 and gives nothing back, so it removes 1/37 of the value from every bet you can place, no matter how many numbers that bet covers. Cover one number or cover eighteen: the arithmetic reaches the same place, because the same pocket is missing from the payout in both cases.

The full payout schedule

Inside bets sit on the numbers themselves; outside bets sit in the boxes around them. The grouping matters for table limits, which are usually set separately for each, and rarely for anything else.

The returns column is total return including your stake, which is how a table displays it and how most disputes about "35 to 1" start. A winning £10 straight-up bet hands back £360: £350 in winnings plus the £10 you had on the felt.

Name the bet before you price it

The schedule above is the easy half. What catches people out at a real table is which shapes the felt will actually accept: a corner is four numbers meeting at one point, not any four numbers you happen to like. Click the layout below and it will name whatever you select and price it. Pick something the felt does not price, four scattered numbers say, and it will tell you that instead, which is the more useful answer.

Every figure is worked on a fixed £10 total rather than £10 a chip. That is the point: ten pounds on a six line and ten pounds spread across those same six numbers both hand back £60. Nothing here is wagered or settled. It is a calculator with a felt for a keypad.

Where variance separates bets that cost the same

If every bet costs 2.70%, the choice between them comes down entirely to the shape of the ride. Betting red for an hour produces a slow drift with frequent small swings. Betting one number for an hour produces long stretches of nothing punctuated by a payout worth 35 spins.

Over 100 spins at £10, an even-money bettor will usually finish somewhere within a couple of hundred pounds of even. A straight-up bettor over the same 100 spins might hit their number five times and finish well ahead, or not hit it at all and be down the lot. Both have an expected loss of £27. Neither is a better bet; they are different products sold at the same price.

This is worth knowing because it is the honest answer to "which bet should I place". Pick the variance you actually want to sit through. Anyone telling you a particular bet is due, or hot, or the smart play, is describing a pattern in independent events, which is the oldest mistake in the game.

The spread behind the average, in real pounds

"Wildly different ride" is a fair description of the gap between a straight-up bet and a bet on red, but it deserves a number rather than just an image. Variance measures how far outcomes typically land from the average, and it works by squaring each result before averaging: a big number squared dwarfs a small one, so a rare, large payout counts for far more than a frequent, small one. Try it on a £1 straight-up bet. Winning £35 once in 37 spins contributes (1/37) × 35² = 1225/37 to the sum; losing £1 the other 36 times contributes (36/37) × 1² = 36/37. Add them and the total is 1261/37, about 34.08. Subtract the expected value squared, a rounding error of roughly 0.0007, and the variance is effectively 34.08, so the standard deviation, its square root, comes to about £5.84 per pound staked. Run the same steps on red, winning £1 eighteen times in 37 and losing £1 the other nineteen, and every squared outcome is exactly 1 regardless of which happens, so the variance is almost exactly 1 and the standard deviation is about £1.00 per pound staked. Same 2.70p expected loss in the pound, but the spread behind it is £5.84 against £1.00.

A dozen or column bet sits between those two extremes, and it is worth working through for that reason alone. It covers twelve numbers at 2 to 1, so a win contributes (12/37) × 2² = 48/37 and a loss contributes (25/37) × 1² = 25/37; the two together come to 73/37, about 1.97. Subtract the same rounding error and the standard deviation is about £1.40 per pound staked, closer to red's £1.00 than to a straight-up number's £5.84. Covering twelve numbers instead of one shrinks the size of the swings without moving the destination: the maths behind every one of these bets still nets out to the identical −2.70%. Variance rises and falls with how concentrated a bet is, in other words, while the edge itself does not move at all.

Scale the two extremes up to the session used earlier in this guide: £10 a spin, 100 spins, £27 of average loss either way. The per-spin standard deviation becomes £58.40 for the straight-up bettor and £10.00 for the red bettor, the per-pound figures multiplied by the £10 stake. Each spin is independent of the last, so the standard deviation of the total across 100 of them grows with the square root of 100, not with 100 itself. Multiply each per-spin figure by 10 and the session-wide spread is £584 for the straight-up player and £100 for the red player, on top of the identical £27 expected cost. One of those two people is routinely a few hundred pounds either side of that £27; the other rarely strays outside £100 of it.

A jumpier bet is more likely to finish ahead, and more likely to lose the lot

That spread cuts in a direction most players would not guess. Work out how often each bettor actually finishes the 100-spin, £10-a-spin session in profit, and the straight-up player comes out ahead more often, not less, despite carrying the identical average loss. Landing the number three times out of 100 is enough: three hits return £1,080 in total against the £1,000 staked across the session, a net profit of £80, and the probability of at least three hits at 1-in-37 odds works out to roughly 51%. Landing on red needs winning at least 51 of the 100 spins to clear the same £1,000, and at a 48.65% chance of red on any given spin, that comes out at around 36%. The wide spread on the straight-up bet pushes more of its outcome distribution past the break-even line on the upside, even though the distribution as a whole still sits below it.

The same spread charges for that advantage somewhere else. Landing the straight-up number zero times in 100 spins happens about 6.5% of the time, and it means losing the full £1,000 staked over the session. A red bettor losing every one of 100 consecutive spins is not a realistic scenario worth quoting odds for; it would take a run long enough to suggest something was wrong with the wheel. A bankroll set aside for a straight-up session has to plan for a real chance of losing the lot, not just the average £27. Betting red never asks that question of a bankroll in the first place.

Why the edge always wins in the end: the spread shrinks, the average does not

That same square-root relationship is the actual mechanism behind "in the long run," a phrase used loosely earlier on this page. Standard deviation across independent spins grows with the square root of the spin count; the average loss grows directly with it. The ratio between the two shrinks every time a session runs longer. At 100 spins on red, the £100 spread sits at roughly 3.7 times the £27 average loss, which is why a session that short can plausibly finish either way. Run the identical bet 10,000 times at the same £10 stake: the average loss scales up in step, to £2,703, while the spread scales up by only the square root of 100, ten times rather than a hundred, to about £1,000. The spread has gone from 3.7 times the average loss down to well under half of it. Nothing about the bet has changed. The session has simply run long enough for the average to outgrow the noise sitting around it.

The one bet that breaks the rule

On an American wheel, the five-number bet covers 0, 00, 1, 2 and 3 for a payout of 6 to 1. Run it through: five pockets win out of 38, so the expectation is (5/38 × 6) − (33/38 × 1), which comes to −3/38, or 7.89%. Every other bet on that table costs 5.26%.

It exists because 36 divided by 5 is not a whole number, so the schedule cannot price it the way it prices everything else, and the rounding went the casino's way. It is the only genuinely stupid bet in a game that is otherwise priced with unusual consistency, and it survives on tables mainly because so few people work it out.

There is no reason to place it.

Call bets and the racetrack

Many live tables add a racetrack overlay for bets defined by position on the wheel rather than on the felt. Voisins du Zéro covers seventeen numbers around the zero, Tiers du Cylindre covers twelve opposite it, and Orphelins covers the eight left over. Neighbour bets cover a chosen number plus those either side of it.

None of these are new bets. Each is a bundle of straight-ups and splits placed in one action, and each therefore costs the same 2.70%. They are a convenience for players who think in wheel sectors instead of grid positions, and they are worth using for that reason alone. If you want to see how the sectors map onto the wheel, the free wheel simulator lays the sequence out without any wagering involved.

"35 to 1" and "36 for 1" are the same price

Two ways of quoting the same payout circulate, and the difference between them is one word. Odds quoted to one describe what you win alongside your returned stake: a winning straight-up at 35 to 1 hands back 36 chips in total, being your original chip plus 35. Odds quoted for one describe the whole return with the stake already inside it, so 36 for 1 also hands back 36 chips. Identical bet, identical money.

The distinction is worth carrying because it is not always roulette you are reading. Slot paytables and some side bets quote for one, and a table that quotes 36 for 1 next to a game quoting 35 to 1 can look like the better deal to anyone skimming. It is not. If a payout is ever quoted without the preposition, assume to one and check before staking anything you would mind losing.

Return to player is the house edge upside down

Online lobbies increasingly label roulette with an RTP figure rather than a house edge, and the two are the same measurement facing opposite directions. A single-zero wheel returns 97.30% and keeps 2.70%. A double-zero wheel returns 94.74% and keeps 5.26%. French rules on the even-money bets push the return on those bets alone up to 98.65%.

One caution about reading RTP on a roulette table, though. On slots the figure is a design choice the studio sets and can change between versions of the same game. On roulette it is arithmetic, fixed entirely by how many pockets are on the wheel, and no operator can adjust it without changing the wheel itself. A roulette table advertising an unusually generous RTP is either counting the French rules, or it is not describing roulette.