Skip to content

21+ Independent live-dealer coverage. The casinos we review are offshore, not US state-licensed. How we review

Safer gambling
Live Casino Reports
Menu

Guide

Optimal Bet Sizing, Expected Value and Odds in Live Roulette

How to calculate roulette odds and expected value, what the Kelly criterion says about bet size (bet zero), and worked maths on why no staking plan wins.

A single-zero roulette wheel spinning in a live casino studio

No bet size can turn roulette into a winning game. On a European single-zero wheel every bet returns $36 for every $37 staked over the long run, a 2.70% house edge, and on an American double-zero wheel the edge is 5.26%. The Kelly criterion, the best-known formula for optimal bet sizing, agrees: plug roulette’s odds into it and the answer is a negative number, which means the optimal stake is zero. This page shows how to calculate the probability and expected value of every roulette bet, how the Kelly formula really works, and what bet sizing can and cannot change.

Updated October 2026. We rewrote the old version of this page, which printed the Kelly formula incorrectly, got two of its probabilities wrong and claimed that Kelly-style betting on red could lead to “millions in winnings”. Every figure below is calculated from the standard payout table and matches the house edges in our live roulette guide.

Calculating Probabilities in Live Roulette

Roulette probability is simple counting. A European wheel has 37 pockets (0 and 1 to 36), an American wheel has 38 (0, 00 and 1 to 36), and on a fair wheel each pocket is equally likely. The probability of a bet winning is the number of pockets it covers divided by the number of pockets on the wheel. Whether you play at a live dealer table, an auto-wheel or a land-based casino makes no difference to that count.

  • Any single number: 1/37 = 2.70% (European) or 1/38 = 2.63% (American).
  • Red: there are 18 red pockets, so 18/37 = 48.65% or 18/38 = 47.37%. That corrects the old version of this page, which said 18/36 and so ignored the zero, and the zero is exactly where the house edge comes from.
  • Not red (black or green): 18 black pockets plus the zero make 19, so 19/37 = 51.35% on a single-zero wheel. Our old page said 20/37, true only with two green pockets; on an American wheel the figure is 20/38 = 52.63%.
  • A dozen or a column: 12/37 = 32.43% or 12/38 = 31.58%.

Below, all the standard bets are brought together. The house edge column is the cost per dollar staked. Next to it, the standard deviation column, which we calculated per one-unit bet on a single-zero wheel, measures how wild the results are, and it becomes important once we get to bet sizing.

Bet Numbers Payout Win chance (European) Win chance (American) House edge (European / American) Standard deviation per unit (European)
Straight up 1 35:1 2.70% 2.63% 2.70% / 5.26% 5.84
Split 2 17:1 5.41% 5.26% 2.70% / 5.26% 4.07
Street 3 11:1 8.11% 7.89% 2.70% / 5.26% 3.28
Corner 4 8:1 10.81% 10.53% 2.70% / 5.26% 2.80
Six line (double street) 6 5:1 16.22% 15.79% 2.70% / 5.26% 2.21
Dozen or column 12 2:1 32.43% 31.58% 2.70% / 5.26% 1.40
Red/black, odd/even, 1-18/19-36 18 1:1 48.65% 47.37% 2.70% / 5.26% 1.00
Top Line (0, 00, 1, 2, 3), American only 5 6:1 n/a 13.16% n/a / 7.89% n/a

Two patterns stand out. The house edge is identical on every standard bet on the same wheel, and what changes from row to row is the volatility: a straight-up bet swings almost six times as hard as a bet on red. Bet size multiplies both the expected loss and the swings, which is why the rest of this page keeps coming back to these two columns. Our page on the house edge in roulette compares these figures with other casino games.

What Is a Random Variable and Expected Value in Roulette?

In probability terms, the net result of a roulette bet is a discrete random variable: it can take only a few fixed values, each with a known probability. Bet $10 on red and the result is +$10 (the ball lands on red) or -$10 (it does not). There is nothing in between.

Expected value (EV) is the probability-weighted average of those outcomes. The formula is EV = (probability of winning × net win) – (probability of losing × stake lost). For the $10 red bet on a European wheel:

  • EV = (18/37 × $10) – (19/37 × $10) = $180/37 – $190/37 = -$10/37 = -$0.27.
  • On an American wheel: (18/38 × $10) – (20/38 × $10) = -$20/38 = -$0.53.

Now try a $10 straight-up bet on a single number. A win pays 35 × $10 = $350, so EV = (1/37 × $350) – (36/37 × $10) = $350/37 – $360/37 = -$10/37, or -$0.27 again. A completely different bet gives exactly the same expected loss. Divide -$0.27 by the $10 stake and you get -2.70%, which is the house edge; on the American wheel the same sum gives -5.26%.

EV is a long-run average, not a forecast for one session. You will never lose exactly 27 cents on a single spin: you either win $10 or lose $10. Over thousands of spins, though, the average result per spin closes in on the EV, and that is the force every bet-sizing plan has to fight. Seen from the player’s side, the same idea is the return to player figure explained in our RTP guide: a 2.70% house edge is a 97.30% RTP.

Roulette Payouts

Roulette payouts are quoted as “x to 1”: you win x for every 1 you stake and keep your original chip. So a straight-up bet paying 35 to 1 returns $36 in total for a $1 chip, $35 of winnings plus the $1 stake. Payouts at a live table are settled automatically, but placement still matters: a chip that sits on a line is a split, not a straight-up bet, so check the bet slip before the betting window closes.

The house edge hides in the gap between the payout and the true odds. A single number on a European wheel has true odds of 36 to 1 against (36 losing pockets, 1 winning), but the casino pays 35 to 1. A split has true odds of 35 to 2, which is 17.5 to 1, and pays 17 to 1. Every standard bet is shaved by the same proportion, which is why they all share the same edge.

Odds and Payouts on Inside Bets

Inside bets go on single numbers or small groups. They win less often and pay more, but the trade is priced exactly, so higher risk does not buy a better return:

  • Straight up: one number, pays 35:1. Wins 1 spin in 37 on average on a single-zero wheel.
  • Split: a chip on the line between two adjacent numbers, pays 17:1.
  • Street: three numbers in a row, with the chip on the outer line of the row, pays 11:1.
  • Corner (square): four numbers that meet at one point, pays 8:1.
  • Six line (double street): two adjoining rows of three, pays 5:1.
  • Top Line (basket, five-number bet): 0, 00, 1, 2 and 3 on an American table, pays 6:1. Wizard of Odds puts its house edge at 7.89%, the worst bet on the layout. Single-zero tables have no Top Line; they offer the first four (0, 1, 2, 3) at 8:1 instead, which carries the normal 2.70%.

The old page drew the wrong lesson from this list. Yes, the payout falls as coverage rises, but the house edge does not move: a corner and a straight-up bet cost the same 2.70% per dollar on a European wheel. Covering more numbers makes you win more often and lose less per losing spin. It does not reduce the price. The worked layouts in our guide to roulette bet combinations show this with real chip spreads.

What Is the Kelly Criterion? How to Calculate Optimal Bet Size

John L. Kelly Jr. published the criterion in 1956 in the Bell System Technical Journal. It answers one question: if you have a real edge, what fraction of your bankroll should you stake on each bet to grow the bankroll as fast as possible over the long run? For a bet that wins with probability p and pays net odds of b to 1, the Kelly fraction is:

f = (b × p – q) / b, where q = 1 – p is the probability of losing.

The old page printed this as “F = bp – (1-p)b”, which is a different expression and gives wrong answers for any odds other than even money. The numerator, b × p – q, is simply the expected profit per unit staked, so the rule reads: stake your edge divided by the odds.

Worked example: a game with an edge

Suppose a hypothetical even-money bet wins 51% of the time. Then b = 1, p = 0.51 and q = 0.49, so f = (0.51 – 0.49) / 1 = 0.02. Kelly says stake 2% of the current bankroll on each bet, so $20 from a $1,000 bankroll, re-calculated after every result. If the win rate were 53%, the fraction would be 6%. These were the examples on the old page, and they are correct, but notice what they assume: a bet that wins more often than it loses at even money. No roulette bet does that.

The odds matter too. A bet paying 2 to 1 that wins 40% of the time has f = (2 × 0.40 – 0.60) / 2 = 0.10, or 10% of bankroll. The edge there is 20 cents per dollar staked, but the bet loses more often than it wins, so Kelly scales the stake down by the odds.

Worked example: roulette

Run the same formula on a $1 bet on red at a European table: b = 1, p = 18/37 and q = 19/37. Then f = (18/37 – 19/37) / 1 = -1/37 = -2.70%. For a straight-up bet, b = 35, p = 1/37 and q = 36/37, so f = (35/37 – 36/37) / 35 = -0.08%. On an American wheel, red gives -5.26%.

A negative Kelly fraction means the formula wants you on the other side of the bet, which only the casino can take. For the player the optimal stake is zero. Kelly does not offer a “best” way to bet on roulette; it proves there is none. The old version of this page claimed that live dealer roulette gives “the best results” with Kelly and that sticking to red or black makes the method work. Neither is true: red and black carry the same 2.70% edge as every other standard bet.

Over-betting, Under-betting and Fractional Kelly

When an edge does exist, Kelly’s real lesson is that stake size has a sweet spot. Bet too little and the bankroll grows slowly; bet too much and the swings drag the growth rate down until it turns negative. The table uses the 51% even-money bet from above and shows the expected growth of the bankroll per bet (the average of the logarithm of the result) and what that means after 1,000 bets.

Fraction of bankroll per bet Relation to Kelly Expected growth per bet Typical bankroll multiple after 1,000 bets
1% Half Kelly +0.015% 1.16x
2% Full Kelly +0.020% 1.22x
3% 1.5x Kelly +0.015% 1.16x
4% Double Kelly about 0% 1.00x
5% 2.5x Kelly -0.025% 0.78x
10% 5x Kelly -0.30% 0.05x

Three things follow, and they match what Wizard of Odds states on its Kelly criterion page. Full Kelly gives the fastest growth. Double Kelly gives zero expected growth even though every single bet still has a positive expected value, and anything beyond that shrinks the bankroll over time. Half Kelly keeps three quarters of the growth (0.015% against 0.020% per bet) while cutting the volatility in half, which is why advantage players often bet a fraction of Kelly: their estimate of the edge is never exact.

That last point also corrects the old page, which warned against “under-betting” and described flat bets of the same amount as a mistake. In a game with an edge, betting below Kelly costs some growth but buys a lot of safety. In a game without an edge, such as roulette, the smaller the total amount you put through the wheel, the less you lose on average. Under-betting is the cheapest thing you can do.

What Bet Size Changes in a Negative-expectation Game

If no bet size beats roulette, does sizing matter at all? Yes, in two ways that have nothing to do with “winning in the long run”.

1. Your expected loss scales with total action

Expected loss = total amount wagered × house edge. One $100 bet on red at a European table costs $2.70 on average. Ten $10 bets cost the same $2.70, and so do one hundred $1 bets. Most players do not stop after a fixed amount of action, though: they play for a fixed time. At $10 a spin and 50 spins an hour, inside the 40 to 60 spins an hour that a croupier-run table typically manages, you put $500 through the wheel and the average cost is $13.51 an hour on a European wheel or $26.32 on an American one. Faster formats such as auto-wheels raise the hourly cost at the same stake.

2. Small bets make reaching a target less likely, not more

Say you have $100 and want to finish with $200. The classic gambler’s ruin formula gives the probability of reaching the target before going broke when betting a fixed amount on red each spin. Bold play, a single $100 bet, is the most likely route to a target in a game with a house edge.

Strategy to turn $100 into $200 European wheel American wheel
One $100 bet on red 48.65% 47.37%
$10 bets on red until you reach $200 or $0 36.8% 25.9%
$5 bets on red 25.3% 10.8%
$2 bets on red 6.3% 0.5%
$1 bets on red 0.45% 0.003%

The longer you play, the more the house edge grinds through. That is the opposite of what the old page suggested, and it is why small, careful bets are a good way to make a fixed budget last for entertainment and a poor way to try to double it. Neither approach changes the 2.70% or 5.26% that every dollar costs.

How often you are still ahead

Short sessions are noisy, so plenty of players finish ahead. The odds of that fade with every spin. Here is our exact binomial calculation for a player betting the same amount on red each spin:

Spins played Chance of being ahead (European) Chance of being ahead (American)
10 34.4% (another 24.5% exactly even) 31.4% (24.3% even)
100 35.5% (7.7% even) 26.5% (6.9% even)
1,000 18.8% 4.5%
10,000 about 0.3% under 0.0001%

Staking Systems: Same Edge, Different Shape

Progressions such as the Martingale, the reverse Martingale, Fibonacci and d’Alembert are bet-sizing rules too. None changes the house edge, because each bet in the sequence is still a separate bet at 2.70% or 5.26%. A quick check with the Martingale on European red:

  • Start at $10 and double after each loss, allowing up to seven bets ($10, $20, $40, $80, $160, $320, $640). Covering the full run of seven takes $1,270.
  • Losing all seven has a probability of 19/37 raised to the 7th power, about 0.94%, or roughly once every 106 sequences, the same figure our roulette hub uses.
  • Each sequence wins $10 with probability 99.06% and loses $1,270 with probability 0.94%. EV per sequence = 0.9906 × $10 – 0.0094 × $1,270, about -$2.05.
  • The average amount staked per sequence is $75.94, and -$2.05 / $75.94 = -2.70%. That is the house edge, to the decimal.

The Martingale wins small amounts often and loses a large amount rarely; the reverse Martingale does the opposite. Both land on the same average. More myths of this kind are taken apart on our roulette myths page.

A Practical Way to Size Roulette Bets

Since the mathematically optimal stake is zero, the useful question is how to size bets for entertainment at a cost you choose in advance:

  1. Pick the cheapest wheel. A single-zero table halves the cost of an American one at the same stake. French tables with La Partage or En Prison cut the even-money edge to 1.35%, per our roulette hub.
  2. Set a session budget you are willing to lose in full, separate from money for bills and savings.
  3. Work out the expected cost before you sit down. Multiply stake, spins per hour and house edge. If $13.51 an hour for $10 European spins is too much, lower the stake or play fewer spins.
  4. Choose the stake from the volatility you want. A budget of 50 to 100 units is a common rule of thumb for even-money play; straight-up bets, with almost six times the standard deviation, empty a small budget much faster on a bad run.
  5. Use loss and time limits that the casino lets you set, and stop when you hit them rather than raising the stake to “win it back”.

Bonuses rarely help here. Bovada and Ignition count live dealer games at 0% toward bonus wagering and software roulette at 5%, and BetOnline excludes roulette from its casino bonuses, according to each operator’s bonus terms.

The Odds of the Same Number Being Spun Several Times in a Row

Every spin is independent, so the chance that the next spin repeats the last number is always 1 in 37 on a European wheel (1 in 38 on an American one), no matter what came before. The odds get long only when you name the streak in advance:

  • Whatever number lands first being repeated on the next two spins: 1/37 × 1/37, about 1 in 1,369 on a single-zero wheel. If you name the number before the first spin, all three spins must hit it: 1/37 cubed, about 1 in 50,653.
  • A number you choose hitting seven times running: 1 in 37 to the 7th power, about 1 in 94.9 billion on a European wheel and about 1 in 114.4 billion on an American wheel.
  • Letting a $1 straight-up bet ride for seven straight wins returns 36 to the 7th power, $78,364,164,096 in total. Our old page gave $64,339,296,875, which is 35 to the 7th power and leaves out the returned stake that gets re-bet each time. No table maximum would allow either.

The best-documented streak is from Monte Carlo on 18 August 1913, when black came up 26 times in a row. Wikipedia’s article on the gambler’s fallacy puts the chance of either colour running 26 times on a single-zero wheel at about 1 in 68.4 million, and gamblers lost heavily betting on red because they believed it was “due”. The old page also repeated a story about the number 19 hitting seven times in a row at the Rio in Las Vegas; we could not find a primary record of it, so it is not repeated here.

Where to Play Live Roulette

Eight states have legalised online casinos; seven markets are live (New Jersey, Pennsylvania, Michigan, West Virginia, Connecticut, Delaware and Rhode Island), and Maine legalised a tribal-only market (LD 1164: became law without the governor’s signature in January 2026, effective 29 July 2026) that has not launched as of October 2026. Players in those seven states can find live single-zero tables at state-licensed casinos. Our live roulette hub covers the variants and the casinos that accept US players, and our responsible gambling page covers deposit limits and self-exclusion. If gambling stops being fun, call the National Problem Gambling Helpline at 1-800-MY-RESET (1-800-697-3738).

Frequently Asked Questions

What is the optimal bet size in roulette?

Zero, in strict mathematical terms. The Kelly criterion gives -2.70% for red on a European wheel and -5.26% on an American wheel, and a negative Kelly fraction means no bet has positive expected value. In practice, choose a stake whose expected cost (stake × spins × house edge) fits your entertainment budget.

Can the Kelly criterion be used in roulette?

Only to show that you should not bet. Kelly assumes an edge; every standard roulette bet has a negative edge, so the formula returns a negative fraction for all of them. It is useful in games or situations where a player has a measurable edge.

How do you calculate expected value in roulette?

Multiply each outcome by its probability and add them. For $10 on red at a European table: 18/37 × $10 – 19/37 × $10 = -$0.27, or -2.70% of the stake. The same calculation gives -2.70% for every standard bet on that wheel and -5.26% on an American wheel.

Does betting more or less change the house edge?

No. The house edge is a percentage of each bet, so it stays at 2.70% or 5.26% whatever the stake. Bet size changes your expected loss in dollars and how much your results swing, not the percentage.

Is it better to make one big bet or many small bets in roulette?

If the goal is to reach a target, one big bet is more likely to succeed: one $100 bet on red doubles $100 48.65% of the time on a European wheel, while $1 bets reach $200 only about 0.45% of the time. If the goal is to make a budget last, small bets do that, at the same cost per dollar wagered.

What are the odds of the same number coming up twice in a row?

The chance that the next spin repeats the previous number is 1 in 37 on a European wheel and 1 in 38 on an American wheel. Past spins have no effect on the next one.

About the author

Live Casino Reports Editorial Team

The Live Casino Reports Editorial Team researches and edits the site's US casino reviews, bonus pages, game hubs and trust pages against regulator, operator and studio sources, with a named second reviewer on money pages.

Fact-checked by Tuah bin Razak.