D’Alembert System in Roulette: How Often It Wins and What It Costs

The D’Alembert system is the calmest of the classic roulette systems. You bet on red or black, add one unit after a loss and take one unit off after a win. The stakes never jump, and the idea behind it sounds fair: wins and losses should even out over time, and when they do, you are ahead. We simulated one million sessions to see what happens on a real wheel with a zero.

The short answer: with a bankroll of 200 units and a target of +50, the D’Alembert system ended 67% of sessions in profit, but it lost the whole bankroll in 28%. That is more than the Martingale (16%) and more than twice the Fibonacci (12%). It also put the most money on the table of all ten systems we tested and lost the most per session. Like every other way of betting on roulette, it lost 2.7% of what it staked. Try your own settings in the simulator right below.

67%
Of sessions ended in profit (bankroll 200, target +50)
28%
Of sessions lost the whole bankroll
1,050
Units staked per session on average (Martingale: 538)
2.7%
Lost on average per unit staked

1. Try the D’Alembert system in the simulator

The simulator starts with the D’Alembert system and our standard settings. Change the bankroll, the base bet, the table limit or the target and see how the share of good and bad sessions moves. Switch to any other system to compare them on the same settings.

2. How the D’Alembert system works

You bet on an even-money chance: red or black, odd or even, or low or high (1 to 18, 19 to 36). All three pay 1:1 and lose on zero. Then you follow three rules:

  • Start with 1 unit.
  • After a loss, add 1 unit to your next bet.
  • After a win, take 1 unit off your next bet, but never go below 1 unit.
SpinStakeResultRunning totalNext stake
11Loss−12
22Loss−33
33Loss−64
44Win−23
53Win+12
62Win+31
Three losses and three wins with the D’Alembert. Each win is placed at a stake one unit higher than the loss it answers.

This is the idea that makes the system popular. After three losses and three wins you are 3 units ahead, because every win comes at a stake one unit higher than the matching loss. As long as you win as often as you lose, you gain one unit for every win, whatever the order.

The problem: wins and losses do not even out

On a single-zero wheel you lose an even-money bet 19 times in 37 and win it 18 times. Over a long evening you will usually lose more bets than you win, and every loss above the number of wins pushes your stake up by one unit for good. Those extra losses get more and more expensive: with ten more losses than wins, they cost 1 + 2 + … + 10 = 55 units, and you only get one unit back for each win. With about twenty more losses than wins, the extra losses add up to around 200 units, and a 200-unit bankroll is gone.

The bets themselves stay small. In our standard test the largest stake of a session was 15 units in the middle case, and in nine sessions out of ten it stayed at 22 units or less. The table maximum never stopped the D’Alembert, even with a maximum of 50 units the results were exactly the same. It loses its money slowly, through many medium-sized bets after a run in which losses lead.

3. What our simulation shows

Our standard test is the same for every system: bankroll 200 units (for example $1,000 with $5 bets), table maximum 100 units, bets on red, zero loses, and the session ends at +50 units, when the money runs out, or after 300 spins. We played this one million times.

40 sessions, spin by spin

Result after each spin in 40 D’Alembert sessions
Reached +50 (27 of 40)Stopped after 300 spins (1 of 40)Lost the bankroll (12 of 40)
40 D'Alembert sessions: result after each spin −200 −150 −100 −50 0 +50 0 50 100 150 200 250 300 Session 1: ended +55 units after 197 spins Session 2: ended +50 units after 108 spins Session 4: ended +50 units after 97 spins Session 5: ended +53 units after 141 spins Session 6: ended +51 units after 101 spins Session 9: ended +54 units after 123 spins Session 10: ended +55 units after 133 spins Session 12: ended +54 units after 132 spins Session 13: ended +50 units after 94 spins Session 14: ended +51 units after 113 spins Session 15: ended +56 units after 156 spins Session 18: ended +51 units after 170 spins Session 20: ended +50 units after 102 spins Session 21: ended +51 units after 102 spins Session 23: ended +50 units after 103 spins Session 24: ended +53 units after 120 spins Session 25: ended +54 units after 204 spins Session 26: ended +56 units after 224 spins Session 27: ended −103 units after 300 spins Session 28: ended +52 units after 145 spins Session 30: ended +50 units after 123 spins Session 31: ended +50 units after 89 spins Session 32: ended +50 units after 99 spins Session 33: ended +50 units after 82 spins Session 34: ended +52 units after 146 spins Session 35: ended +51 units after 104 spins Session 38: ended +50 units after 112 spins Session 40: ended +50 units after 90 spins Session 3: ended −200 units after 299 spins Session 7: ended −200 units after 133 spins Session 8: ended −200 units after 182 spins Session 11: ended −200 units after 110 spins Session 16: ended −200 units after 154 spins Session 17: ended −200 units after 153 spins Session 19: ended −200 units after 49 spins Session 22: ended −200 units after 121 spins Session 29: ended −200 units after 89 spins Session 36: ended −200 units after 200 spins Session 37: ended −200 units after 155 spins Session 39: ended −200 units after 223 spins Spin Result (units)
The first 40 of one million simulated sessions in the standard test, not selected. Blue: target reached. Grey: stopped after 300 spins. Orange: bankroll lost. Hover over a line for its result. Source: GamblingSociety simulation.

The picture is very different from the Martingale. There are no sudden drops. Instead, some lines turn downward and keep sliding, because each further loss is placed at a higher stake than the last. Of these 40 sessions, 27 reached the target, 12 lost the bankroll and 1 was still going after 300 spins. Over all one million sessions, two thirds were decided within the first 150 spins.

D’Alembert against Martingale and Fibonacci

D’Alembert compared in the same test
D’AlembertMartingaleFibonacci
D'Alembert compared in the same test 0% 25% 50% 75% 100% Target +50 reached: D'Alembert 67.0%, Martingale 62.3%, Fibonacci 48.2% 67% 62% 48% Target +50 reached Ended ahead: D'Alembert 67.1%, Martingale 71.1%, Fibonacci 65.9% 67% 71% 66% Ended ahead Whole bankroll lost: D'Alembert 28.4%, Martingale 16.2%, Fibonacci 11.6% 28% 16% 12% Whole bankroll lost Bankroll 200 units, table maximum 100, target +50, at most 300 spins
All three systems with a bankroll of 200 units, table maximum 100, target +50, at most 300 spins, one million sessions each. Hover over a group for the numbers. Source: GamblingSociety simulation.

The D’Alembert system reached the target more often than both (67%, against 62% for the Martingale and 48% for the Fibonacci), and it needed a median of 110 spins to get there. But it also lost the whole bankroll in 28% of sessions. Over all sessions it staked 1,050 units on average, almost twice as much as the Martingale (538), and at 2.7% per unit staked, that made it the most expensive system in our test: −28.4 units per session, against −14.6 for the Martingale and −19.4 for the Fibonacci.

The size of the target

Target reached and bankroll lost, by target size
Target reachedWhole bankroll lost
D'Alembert: target reached and bankroll lost, by target size 0% 25% 50% 75% 100% Target +10 units: reached in 91.2%, ended ahead in 91.2%, whole bankroll lost in 8.1%, average result −7.1 units 91% 8% +10 units Target +20 units: reached in 84.0%, ended ahead in 84.0%, whole bankroll lost in 14.5%, average result −13.2 units 84% 15% +20 units Target +30 units: reached in 77.7%, ended ahead in 77.7%, whole bankroll lost in 20.0%, average result −18.6 units 78% 20% +30 units Target +50 units: reached in 67.0%, ended ahead in 67.1%, whole bankroll lost in 28.4%, average result −28.4 units 67% 28% +50 units Target +75 units: reached in 55.6%, ended ahead in 56.4%, whole bankroll lost in 34.9%, average result −38.7 units 56% 35% +75 units Win target (bankroll 200 units, at most 300 spins)
D’Alembert with a bankroll of 200 units and a table maximum of 100 units, at most 300 spins, one million sessions per target. Hover over a group for the numbers. Source: GamblingSociety simulation.

With +10 units, 91% of sessions reached the target and 8% lost the bankroll. With +20 it was 84% and 15%, and with +75 only 56% and 35%. Every extra unit of target means more spins, and more spins give the losses more time to pull ahead of the wins.

All scenarios

ScenarioTarget reachedSession ended aheadWhole bankroll lostAverage result (units)Lost per 100 staked
Standard: bankroll 200, table max 100, target +5067%67%28%−28.42.7
Table max 5067%67%28%−28.42.7
La Partage (half back on zero)73%73%23%−13.21.4
Stop at a loss of 100 units56%56%0%−18.12.7
Target +20, stop at a loss of 100 units77%77%0%−8.42.7
Bankroll 50, target +1274%74%26%−3.52.7
Bankroll 100, target +2571%71%28%−10.32.7
Bankroll 500, target +12534%52%11%−79.52.7
Target +1091%91%8%−7.12.7
Target +2084%84%15%−13.22.7
Target +3078%78%20%−18.62.7
Target +7556%56%35%−38.72.7
One million sessions per scenario, at most 300 spins. Everything not mentioned is as in the standard test. A stop at a loss of 100 units can end a little lower, because the last bet may take the loss past 100. Source: GamblingSociety simulation.

Two lines stand out. With a bankroll of 500 units and a target of +125, the average session lost 79.5 units, almost three times as much as in the standard test, because a bigger bankroll lets the stakes climb higher before the session ends (on average to 24 units instead of 15). A stop-loss at half the bankroll removed the lost bankrolls completely and cut the average loss to 18.1 units, but 44% of sessions ended at that stop. La Partage halves the price, as with every system: 13.2 units lost per session on average.

4. Variants of the D’Alembert system

  • D’Alembert with a stop-loss. You stop the session when you are 100 units down. Together with a target of +20, 77% of our sessions reached the target and the average session cost 8.4 units instead of 28.4.
  • Contra D’Alembert. The reverse: one unit more after a win, one unit less after a loss. It bets more while you are winning instead of while you are losing. We have not tested it separately. The cost stays at 2.7% of every stake on red or black.
  • Fibonacci. Also a slow progression, but a win takes the stake two steps back. It lost the bankroll far less often in our test (12%). Our Fibonacci page has the full test.
  • Martingale. Doubling after every loss ends each series with a win of 1 unit. It loses rarely but then heavily. Our Martingale page shows how often.

5. How to play the D’Alembert system sensibly

The D’Alembert system cannot beat the house edge. If you treat it as entertainment with a fixed price, these rules from the numbers above help:

  • Keep the target small. With +10 units, 91% of our sessions reached it. Every extra unit of target means more spins for the losses to pull ahead.
  • Set a stop-loss. The D’Alembert system does not lose its money in one big hit. It slides. A fixed point where you stop, for example at half the bankroll, keeps a slide from taking everything.
  • Watch your stake, not only your balance. When your stake has climbed to 10 or 15 units, losses are leading clearly. That is a good moment to stop or start again at 1 unit with a new budget.
  • Do not raise the bankroll to make it safer. In our test a bankroll of 500 units lost more per session than one of 200, because the stakes had more room to climb.
  • Play at a table with La Partage if you can. It halves the cost of every bet on red or black.
  • Set your budget before the first spin. When it is gone, the evening is over.

Frequently asked questions

Does the D’Alembert system work in roulette?

It wins most sessions, but not in the long run. Over many sessions it loses 2.7% of everything you stake, like every other roulette system. In our standard test it ended 67% of sessions ahead but lost the whole bankroll in 28%, and the average session lost 28.4 units.

Why is it called the D’Alembert?

It is named after Jean le Rond d’Alembert, an 18th-century French mathematician who argued that after a run of one result, the other result becomes more likely. That idea does not hold for roulette: every spin is independent, as our test with real casino spins shows.

Is the D’Alembert system safer than the Martingale?

Each bet is much smaller, so it feels safer. In our test it lost the whole bankroll more often than the Martingale (28% against 16%) and lost more per session on average, because it puts much more money on the table.

How much bankroll do I need for the D’Alembert system?

There is no fixed limit like with the Martingale, because the stakes grow by one unit at a time. In our test the largest stake stayed at 22 units or less in nine sessions out of ten, so a bankroll of 200 base bets covers most sessions. A bigger bankroll did not lower the cost, it made the average loss bigger.

Which bet is best for the D’Alembert?

Red or black, odd or even, and 1 to 18 or 19 to 36 all have the same odds. The table rules matter more: with La Partage you get half your stake back on zero, which halves the house edge on these bets.

Does the D’Alembert system work online?

The odds are the same online and in a casino. Online tables play faster, so you reach your target sooner, but you also make more bets per hour and pay more of the 2.7% edge per hour.

How we tested it

  • Simulator: our own program, one million sessions per scenario. We checked the standard test with a second, independent program (150,000 sessions): it found 67.0% of targets reached and 28.3% lost bankrolls, the same as the main simulation within its precision.
  • Rules: single-zero wheel, bets on red, all even-money bets lose on zero (variant: La Partage). One unit more after a loss, one unit less after a win, never below 1 unit. A session ends at the target, when the money runs out, or after 300 spins.
  • Limits: one player, no mistakes, no side bets. Real tables play roughly 30 to 60 spins per hour, so 300 spins are a long evening.

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.