Roulette strategies

Two families of methods circulate around the roulette wheel, and they fail for different reasons, which is why we prefer to take them apart one at a time. The first changes the size of the bets after each result (the "progressions"), while the second claims to know where the ball will land, from past numbers, from a defect of the wheel, from the dealer's hand or from a measurement of the ball.

What a strategy can change is quickly stated: the shape of the results and the total that crosses the table. What it cannot change is the share the wheel keeps, 2.70 per cent on a single zero (each bet is worked out on our bets page).

Progressions change the total, not the edge

A "progression" changes the stake after each result, and the classic "martingale" doubles after every loss so that the first win recovers everything plus one unit. The expected loss of an evening, however, is the edge multiplied by everything wagered; a system that reorders and resizes the bets can move that sum in one way only, by moving the total, as the general proof on the betting systems page shows.

A player who starts a martingale at $5 on a table whose maximum is $200 can double five times (10, 20, 40, 80, 160) before the table refuses the next stake. Computed rung by rung from the probability of each loss, the average stake comes to about $15.91 a spin, more than three times a "flat" stake of $5. Over 200 spins that is some $3,180 across the table instead of $1,000, and an expected loss of about $86 instead of $27.03!

The slower negative progressions (Fibonacci, D'Alembert) follow the same arithmetic, and the positive ones, which raise after a win, make the opposite trade; both are worked out on the negative progressions page and the positive progressions page. A flat stake is the yardstick, and the Kelly criterion applied to a bet on red returns a negative fraction: the right stake is none.

Hot and cold numbers, and the gambler's fallacy

A board of the latest numbers, with the "hot" ones that have come out often and the "cold" ones that have not, is a familiar sight beside a wheel, and it invites a reading the wheel does not allow. Each spin starts again from the same thirty-seven pockets. Even a perfect wheel produces irregular counts on its own, and in a run of 37 spins about 13 numbers do not come out at all (37 times 36/37 raised to the 37th power).

Over 370 spins, when each number is expected ten times, the most frequent typically appears 17 times and the least frequent 4 (medians of 20,000 sessions we simulated). A case often quoted is the evening of 18 August 1913 at the Monte Carlo casino, when black came up twenty-six times in a row and the players who bet against it, convinced that red was "due", lost millions of francs.

A run of 26 of either colour has a probability of about 1 in 68.4 million on a single-zero wheel (twice 18/37 raised to the 26th power); the twenty-seventh spin was still red with a probability of 18 in 37, exactly as the first one had been!

Believing that a colour or a number is owed to the player is the "gambler's fallacy" (the idea that independent events correct one another). Its mirror image, the faith in a number that keeps coming out, is the basis of systems such as the "Andrucci" (betting on the numbers seen most often over a few hundred spins), which would make sense only on a wheel with a physical defect. That case exists, and it is the subject of the next section.

Wheel bias, real on old wheels

A physical wheel can have a defect (a worn pocket, a loose separator, a rotor slightly out of level) that makes some numbers come out more often than chance allows, and the phenomenon of the "biased wheel" is documented. In the early 1990s the García-Pelayo family recorded several thousand spins at the Casino Gran Madrid and found, by computer, a wheel that favoured certain numbers. They bet on those numbers until the casino barred them; according to accounts of the case, some casinos then began checking their wheels more often.

The method needs a physical wheel, thousands of recorded spins on that same wheel and a casino that neither moves nor services it, and in our view an "RNG" game (one whose results come from a random number generator) offers none of the three. There is no wheel to wear, only a generator that gives every pocket the same chance on every spin, and nothing for a notebook to discover.

The dealer signature

The "dealer signature" is the theory that a croupier who launches the ball with the same speed and the same gesture, spin after spin, makes the sector where it falls predictable from the number where it started. The theory meets the physics of the wheel, where the ball strikes the metal "deflectors" (small obstacles around the track) and bounces among the pocket separators before it settles, so that small differences at the launch become large ones at the landing.

Online it meets a simpler obstacle. On a live table the bets close on a countdown, before the ball has slowed enough for its path to say anything, so whatever a dealer's hand might reveal arrives too late, and we would not pay for a course that promises you otherwise.

Prediction devices: real cases, closed doors

Measuring the ball and the rotor to predict the landing sector ("visual ballistics" by eye, a "roulette computer" with a device) is the other method that has worked on real wheels, and we know of three cases on record. Edward Thorp and Claude Shannon built what is described as the first wearable computer between 1960 and 1961 and took it to Las Vegas in the summer of 1961. A group of students later known as the "Eudaemons" hid a computer in a shoe in the second half of the 1970s.

In March 2004 three players won 1.3 million pounds at the Ritz in London with a method the press attributed to a "laser scanner", before the police closed the case and returned the money (the three accounts are in Adam Kucharski's "The Perfect Bet"). Each case needed a physical ball, a wheel that turned long enough to be measured and bets still open while it turned, and those doors have closed.

In Nevada the use of any device that "projects the outcome of the game" has been unlawful since 1985 (NRS 465.075, a felony from the first offence), an RNG game has no ball to measure, and a live table online closes its bets on a timer. A device sold online as a "roulette predictor" is a commercial promise and nothing more, for reasons set out on our page on technical promises.

Three choices that change the bill

What remains, once the methods are set aside, are three decisions that the arithmetic recognises, and we consider them the only strategy worth the name. The first is the wheel, since a single zero instead of a double nearly halves the price and the French rules halve it again on even-money bets (figures on our roulette page).

The second is the total. Since the expected loss is the edge times everything wagered, a budget fixed in advance for the evening (not a stake per spin) is the figure that limits the cost, and if you decide to send $500 across a single-zero table, you have fixed the expected loss at $13.51 before the first spin.

The third is the moment to stop, which we would advise you to decide before sitting down and not in the middle of a run, because the wheel offers no better moment than any other. The tools of licensed operators that help a player keep to it (deposit limits, time-outs, "reality checks" that show the time and money spent) are described on our responsible gambling page.

Questions

By the Rand Casinos editorial team · Last updated October 2026 · Online since 2007