Roulette Systems Tested: 10 Strategies, 100 Million Sessions

Almost every roulette player has heard of Martingale, and many have tried a system of their own. Most systems are good at one thing: they win small amounts in most sessions. None of them changes the house edge. We wanted exact numbers for both sides, so we simulated ten popular systems over more than 100 million sessions and tested the ideas behind them on almost 400,000 real spins from a German casino table.

The short version: in our standard test, Martingale finished 71% of all sessions in profit, and every single system still lost about 2.7% of everything it staked. How an evening at the table goes depends far more on your target, your bankroll and the moment you stop. You can test any system yourself with the simulator right below.

2.7%
Lost on average per unit staked, in every system we tested
71%
Of Martingale sessions ended in profit
1 in 37
Chance of any single number, also after 200 spins without it
≈ 400,000
Real spins from one casino table checked for patterns

1. Try it yourself: the roulette system simulator

Pick a system, set your bankroll, base bet, table limit and target, and the simulator plays thousands of sessions in your browser. You see how often the target is reached, how often the bankroll is gone and what single sessions look like, spin by spin. The default settings are the standard test we use further down this page.

2. What a roulette system can do, and what it can’t

A roulette system is a rule for how much to bet and on what, usually based on the last results. Most systems raise the stake after a loss, so that one win makes up for several losses: Martingale, Fibonacci, D’Alembert and Labouchère work this way. Paroli raises the stake after a win instead. Other systems wait for a pattern before they bet at all, for example a number that has not come up for a long time.

On a single-zero wheel every bet has the same price. The payouts are calculated as if the wheel had 36 numbers, but it has 37, and that difference costs you 2.7% of your stake on average. A system cannot change this price, because every spin is independent of the ones before. What a system does change is the shape of your results: how often you win, how much, and how big the rare losses are. That shape is what you feel at the table, and it is what our simulation measures.

3. The systems we tested

SystemHow it worksType
MartingaleBet on red or black. Double the bet after every loss and go back to 1 unit after a win.Doubling after losses
Grand MartingaleLike Martingale, but add one extra unit after every loss (1, 3, 7, 15 …).Doubling after losses
FibonacciBet along the sequence 1, 1, 2, 3, 5, 8 …: one step up after a loss, two steps back after a win.Slow progression after losses
D’AlembertRaise the bet by 1 unit after a loss and lower it by 1 unit after a win.Slow progression after losses
LabouchèreWrite down 1-2-3-4 and bet the first plus the last number. After a win cross both out, after a loss add the lost bet to the end. The line is done when all numbers are crossed out.Cancellation line
ParoliStart with 1 unit and double after each win, up to three wins in a row. Then start again with 1 unit.Doubling after wins
Serie 5/8 with doublingSix split bets on the 12 numbers opposite the zero (tiers du cylindre). Double the total stake after every miss.Doubling after losses, 12 numbers
SleepersBet 1 unit on every number that has not come up for 100 spins or more.Waiting for “due” numbers
Wait for 5 in a rowAfter five reds (or five blacks) in a row, bet on the other colour with Martingale until it wins.Waiting for a run to end
Flat bet1 unit on red on every spin, no system at all.Comparison
The even-money systems bet on red. We are adding a guide for each system; the links appear here as the guides go online.

4. The test: same budget, same table, ten systems

Every system played the same situation one million times: a bankroll of 200 units (for example $1,000 with $5 bets), a table maximum of 100 units and a single-zero wheel on which all even-money bets lose on zero, as in our free roulette game. A session ended at a win of 50 units, when the money ran out, or after 300 spins, about as many as one live table produces in a whole evening.

When a progression called for a bet above the table maximum or above the money left, the player took the loss and started again at 1 unit. That is what most players do in practice.

How the sessions ended

How 300-spin sessions end, by system
Session ended aheadEnded behind, money leftWhole bankroll lost
How 300-spin sessions end, by roulette system Wait for 5 in a row, then Martingale: ended ahead 92.0%, ended behind, money left 7.9%, whole bankroll lost 0.0% Wait for 5 in a row, then Martingale 92% Labouchère (1-2-3-4): ended ahead 73.6%, ended behind, money left 0.2%, whole bankroll lost 26.2% Labouchère (1-2-3-4) 74% 26% Serie 5/8 with doubling: ended ahead 72.7%, ended behind, money left 0.2%, whole bankroll lost 27.0% Serie 5/8 with doubling 73% 27% Grand Martingale: ended ahead 72.5%, ended behind, money left 5.7%, whole bankroll lost 21.8% Grand Martingale 73% 22% Martingale: ended ahead 71.1%, ended behind, money left 12.7%, whole bankroll lost 16.2% Martingale 71% 13% 16% Sleepers (absent 100+ spins): ended ahead 67.3%, ended behind, money left 13.6%, whole bankroll lost 19.0% Sleepers (absent 100+ spins) 67% 14% 19% D'Alembert: ended ahead 67.1%, ended behind, money left 4.5%, whole bankroll lost 28.4% D'Alembert 67% 28% Fibonacci: ended ahead 65.9%, ended behind, money left 22.5%, whole bankroll lost 11.6% Fibonacci 66% 22% 12% Paroli (3 wins): ended ahead 33.4%, ended behind, money left 66.6%, whole bankroll lost 0.0% Paroli (3 wins) 33% 67% Flat bet (no system): ended ahead 29.9%, ended behind, money left 70.1%, whole bankroll lost 0.0% Flat bet (no system) 30% 70% 0% 25% 50% 75% 100%
Share of one million simulated sessions per system: bankroll 200 units, table maximum 100 units, target +50 units, at most 300 spins. Hover over a bar for the exact numbers. Source: GamblingSociety simulation.

Read each bar from left to right and you see the trade every system makes. Labouchère, Serie 5/8 and the two Martingales finish about three quarters of all sessions in profit. They pay for it with the sessions that end with the whole bankroll gone: 16% for Martingale and 26% to 28% for Labouchère, Serie 5/8 and D’Alembert. Flat betting and Paroli almost never lose everything, but they also end ahead far less often (30% and 33%).

“Wait for 5 in a row” looks best, with 92% of sessions ahead. The reason is that it hardly plays. Five of the same colour in a row are rare, so in 300 spins it places only a few small series of bets. Most of them end with a small win, and now and then a long run meets the table limit.

The bill is always the same

The average loss follows the amount staked
Average result per session against the average amount staked 0 −10 −20 −30 0 200 400 600 800 1,000 Dashed line: 2.7% of everything staked Flat bet (no system): on average 300 units staked, result −8.1 units Flat bet Martingale: on average 538 units staked, result −14.6 units Martingale Grand Martingale: on average 491 units staked, result −13.2 units Grand Martingale and Paroli Fibonacci: on average 718 units staked, result −19.4 units Fibonacci D'Alembert: on average 1,050 units staked, result −28.4 units D'Alembert Labouchère (1-2-3-4): on average 561 units staked, result −15.2 units Labouchère Paroli (3 wins): on average 492 units staked, result −13.3 units Wait for 5 in a row, then Martingale: on average 63 units staked, result −1.7 units Wait for 5 in a row, then Martingale Sleepers (absent 100+ spins): on average 323 units staked, result −8.9 units Sleepers Serie 5/8 with doubling: on average 272 units staked, result −7.3 units Serie 5/8 with doubling Average amount staked per session (units) Average result (units)
Each dot is one system: average amount staked per session and average result, one million sessions each. Hover over a dot for the numbers. Source: GamblingSociety simulation.

This chart shows the part no system can change. Every dot sits on the same line: about 2.7 units lost for every 100 units staked. D’Alembert loses the most per session for one reason only, because it stakes the most. The waiting system loses the least because it bets the least.

So the fair way to compare systems is the price of an evening. With $5 units, an average Martingale session from our test cost about $73 and a flat-bet session about $41. Some evenings end far better and a few far worse, but the average stays where it is.

SystemTarget +50 reachedSession ended aheadWhole bankroll lostAverage result (units)Average staked (units)Lost per 100 staked
Wait for 5 in a row, then Martingale0%92%0%−1.7632.7
Labouchère (1-2-3-4)74%74%26%−15.25612.7
Serie 5/8 with doubling73%73%27%−7.32722.7
Grand Martingale71%73%22%−13.24912.7
Martingale62%71%16%−14.65382.7
Sleepers (absent 100+ spins)65%67%19%−8.93232.7
D’Alembert67%67%28%−28.410502.7
Fibonacci48%66%12%−19.47182.7
Paroli (3 wins)8%33%0%−13.34922.7
Flat bet (no system)0%30%0%−8.13002.7
One million sessions per system, standard scenario as described above. Source: GamblingSociety simulation.

5. What really changes your chances of a good evening

The choice of system matters less than three decisions you make before the first spin. These are the Martingale numbers; the other progressions move in the same direction.

  • A small target. With a target of +20 units instead of +50, Martingale reached its target in 85% of sessions instead of 62%, and lost the whole bankroll in 9% instead of 16%. With a target of +100 units it was 39% and 20%.
  • The right table. On a table with La Partage you get half of an even-money bet back when zero comes up. That halves the price: Martingale lost 7.1 units per session on average instead of 14.6. If you have the choice, play French roulette with La Partage.
  • The table limit. A low maximum (50 units, five doublings) stops the doubling earlier. Fewer sessions end with the whole bankroll lost (9%), but fewer reach the target as well (51%). A higher limit only helps if you also bring more money, and then the rare loss is bigger.
  • Stopping in time. Every extra spin is another bet at 2.7%. Players who leave after a good run keep their profit, and players who carry on give it back over time. Our Serie 5/8 page shows this with a real example.

Decide on your budget for the evening before you start, and see it as the price of the game. When it is used up, the evening is over. A system makes the good evenings more frequent, and the budget keeps the bad ones small.

6. Does roulette have a memory? A test with real spins

Many systems rest on one idea: after a long run of red, black is due, and a number that has been missing for a long time has to come up soon. To test this we used almost 400,000 spins from one roulette table in a German casino, recorded from 2023 to 2026. We checked the numbers for any pattern a betting system could use and found none.

After a run of the same colour

Does a colour run “have to” end?
Observed (bar = 95% range)Expected if the wheel has no memory
Real table: chance that a colour run continues on the next spin 40% 45% 50% 55% no memory: 48.6% After 1 of the same colour in a row (about 190,000 cases): the run continued 48.7% of the time, 95% range 48.5% to 48.9% 1 After 2 of the same colour in a row (about 94,000 cases): the run continued 48.7% of the time, 95% range 48.4% to 49.0% 2 After 3 of the same colour in a row (about 46,000 cases): the run continued 48.5% of the time, 95% range 48.0% to 48.9% 3 After 4 of the same colour in a row (about 22,000 cases): the run continued 49.2% of the time, 95% range 48.5% to 49.8% 4 After 5 of the same colour in a row (about 11,000 cases): the run continued 48.9% of the time, 95% range 48.0% to 49.9% 5 After 6 of the same colour in a row (about 5,300 cases): the run continued 49.5% of the time, 95% range 48.1% to 50.8% 6 After 7 of the same colour in a row (about 2,600 cases): the run continued 49.3% of the time, 95% range 47.4% to 51.2% 7 After 8 of the same colour in a row (about 1,300 cases): the run continued 48.1% of the time, 95% range 45.4% to 50.9% 8 After 9 of the same colour in a row (about 610 cases): the run continued 46.9% of the time, 95% range 43.0% to 50.9% 9 After 10 of the same colour in a row (about 290 cases): the run continued 47.9% of the time, 95% range 42.1% to 53.7% 10 Same colour already came up this many times in a row
Chance that the next spin shows the same colour again after a run of 1 to 10 reds or blacks (zero ends a run). The vertical bars show the 95% range; they grow because long runs are rare. Source: GamblingSociety analysis of almost 400,000 real spins.

After one red or one black, the next spin showed the same colour again 48.7% of the time. After five of the same colour in a row it was 48.9%, and after ten 47.9%, with a wide range because ten in a row happened only a few hundred times. The wheel does not know what happened before. Systems that wait for a run and then bet against it play the same game at the same price, only less often.

Sleepers: numbers that are “overdue”

Do sleepers wake up?
Observed hit rate (bar = 95% range)Fair wheel
Real table: how often a number hits after a long absence 2.0% 2.5% 3.0% 3.5% 4.0% 1 in 37 = 2.70% Any number: hit 2.70% of the time (about 13,000,000 chances), 95% range 2.69% to 2.71% any Absent 50+ spins: hit 2.69% of the time (about 2,600,000 chances), 95% range 2.67% to 2.71% 50+ Absent 100+ spins: hit 2.68% of the time (about 510,000 chances), 95% range 2.64% to 2.73% 100+ Absent 150+ spins: hit 2.72% of the time (about 93,000 chances), 95% range 2.61% to 2.82% 150+ Absent 200+ spins: hit 2.85% of the time (about 15,000 chances), 95% range 2.58% to 3.11% 200+ Absent 250+ spins: hit 3.04% of the time (about 1,900 chances), 95% range 2.28% to 3.81% 250+ Number absent for at least this many spins (same day)
How often a number came up on the next spin after it had been missing for at least 50 to 250 spins on the same day. The vertical bars show the 95% range. Source: GamblingSociety analysis of almost 400,000 real spins.

Each of the 37 numbers comes up 2.70% of the time on average. A number that had been missing for 100 spins came up 2.68% of the time on the next spin, after 150 spins 2.72%. The last two points look a little higher, but they rest on few cases and their range clearly includes 2.70%. A number that is “overdue” has exactly the same chance as any other.

We also tested a classic rule for the Serie 5/8: wait until the 12 numbers have missed several times in a row, then start. They came up 32.4% of the time after no miss, 32.7% after three misses, 32.1% after six and 32.8% after nine. A fair wheel gives 32.4% every time. Waiting leaves the odds where they are, and the only effect is that you bet less often.

The same goes for the law of the third. Numbers that were missing in a rotation of 37 spins came up 2.70% of the time per spin in the next rotation, exactly the 1 in 37 of any other number. Our law of the third analysis shows the details, including a test of the Plein Triumph system on the same spins.

Frequently asked questions

Is there a roulette system that always wins?

No. Every spin is independent of the last one, and every bet on a single-zero wheel loses 2.7% on average. A system can make small wins frequent, but it pays for them with rare large losses. In our test, all ten systems lost between 2.68 and 2.75 units per 100 staked, which is 2.7 within the precision of the simulation.

Which roulette system is the best?

That depends on what you want from the evening. If you like many small wins, Martingale or Labouchère with a small target. If you want to lose slowly and almost never the whole bankroll at once, flat betting or Paroli. None of them is cheaper in the long run. The cheapest choice is a table with La Partage.

Does the Martingale system work?

It wins most sessions: 71% in our standard test. In 16% of sessions a losing streak met the table limit and the whole bankroll was lost. On average Martingale lost 2.7% of what it staked, like every other system. More on our Martingale page.

Do roulette systems work online?

The maths are the same online and in a casino, for computer roulette and live tables alike. Check the table limit before you play a doubling system: on some online tables the maximum allows only five or six doublings from the minimum bet.

Is it legal to use a roulette system?

Yes. A betting system is only a way of choosing your stakes, and casinos have nothing against it. What is not allowed is using electronic devices to predict where the ball will land.

Do hot or cold numbers exist?

Not in a way you can use. On our real table data, numbers that had been missing for a long time came up as often as all others, and a run of one colour did not make the other colour more likely.

How we tested it

  • Simulator: our own program, written for this guide. One million sessions per system and scenario, more than 100 million sessions in total. We checked it against exact calculations for Martingale, Grand Martingale, D’Alembert and Serie 5/8, and the results matched within the precision of the simulation.
  • Rules: single zero, payouts 35:1 (number), 17:1 (split) and 1:1 (red/black), even-money bets lose on zero (variant: La Partage). Standard session: bankroll 200 units, table maximum 100 units, target +50 units, at most 300 spins. A progression starts again at 1 unit when the next bet would be above the table maximum or above the money left. Sleepers and the waiting system watch 200 spins before the session starts.
  • Real data: almost 400,000 spins from one roulette table in a German casino, January 2023 to September 2026, taken from the casino’s published records. We checked every day against the casino’s own daily statistics. Each casino day counts as its own sequence.
  • Limits: one player, no mistakes, no tips or side bets. The real data come from one table; other tables and casinos can differ a little, but no table can change the price of 2.7%.

Related guides

About the author: GamblingSociety is an independent site run by Andreas, who has operated casino and betting guides since the mid-2000s. No bookmaker or casino owns or pays for our content. More about us.