Niko Tosa is the Croatian-born professional gambler whose work on roulette sits at the rare intersection of physics, observation, and casino history. Most “systems” sold to roulette players reduce to variants of progression betting and produce no long-run edge. Tosa’s work is different — not because it beats the math of a fair wheel, but because it identifies the narrow conditions under which the wheel itself stops being mathematically fair. This guide covers his method, the physics it rests on, and why it does not transfer to online roulette.

Who is Niko Tosa?

Tosa is best known in the gambling world as one of the three people involved in the 2004 Ritz Casino case in London, which generated headlines worldwide when a roulette team walked out with around £1.3 million across two nights. Beyond that case, he is one of a small group of advantage players who treat roulette as a physics problem rather than a probability problem — a tradition that runs back through Gonzalo García-Pelayo in 1990s Madrid and Joseph Jagger in 1873 Monte Carlo, both of whom are profiled in our famous roulette players guide.

What separates Tosa from typical “winning system” sellers is what he doesn’t claim. He doesn’t claim to beat random spins. He doesn’t claim to have found a betting progression that defeats the house edge. The claim is narrower and harder to dismiss: that on certain physical wheels, in certain conditions, the outcome is measurably less random than it should be — and that this can be detected and exploited within the time window between the ball release and the croupier’s call of “no more bets.”

The Method: Visual Ballistics & Sector Prediction

The technique Tosa is associated with is broadly known as visual ballistics or sector prediction. It does not rely on past results — there is no hot-number tracking, no “due” numbers, no patterns in the recent history. The variables observed are physical and happen during the current spin:

• Wheel speed. The rate at which the rotor is turning when the ball is released.
• Ball speed. The initial velocity of the ball on the outer track.
• Release point. The position of the croupier’s hand relative to the wheel at the moment the ball leaves it.
• Deceleration rate. How quickly the ball loses speed on the track before dropping.
• Drop point pattern. Whether the ball tends to leave the track at a consistent rotor position — a sign of wheel tilt or worn frets.

Combined, these variables let a trained observer predict not the exact pocket but a likely sector — typically a contiguous block of 6 to 10 numbers covering roughly a quarter of the wheel. If the prediction is meaningfully better than chance even some of the time, the house edge can be reversed. At European odds of 35:1 on a straight-up, an observer only needs to lift their accuracy from 1-in-37 to roughly 1-in-30 across their target sector to be playing at a positive expected value.

The Physics Behind the Prediction

A roulette wheel is, mechanically, a fairly simple system. A rotor spins in one direction, a ball travels in the opposite direction along a fixed track, friction slows the ball until gravity pulls it inward, it bounces between the frets, and it settles in a pocket. Each step is governed by classical physics. None of it is truly random in the sense that quantum events are random — it is deterministic, dependent on initial conditions that humans can’t measure precisely enough to predict.

That last clause is where advantage play lives. The system is deterministic. Humans can’t usually measure it precisely. But “usually” is doing a lot of work in that sentence.

In 2012, the physicists J. Doyne Farmer and Norman Packard published their famous Eudaemonic Pie work using shoe-mounted computers in 1970s Las Vegas. In 2012, mathematicians Michael Small and Chi Kong Tse published a peer-reviewed paper in the journal Chaos demonstrating mathematically that with sufficient measurement precision, sector prediction on a roulette wheel could yield an expected return of around 18% per spin — vastly above the house edge. The paper did not say roulette is easy to beat. It said the physics permits it under specific conditions.

The conditions are demanding:

• The wheel must be physically observable in real time — no obstructions, no late-bet cutoffs that close betting too early in the spin.
• The wheel must have either a measurable bias or a dealer with a recognisable release pattern.
• The observer must be skilled enough to perform calculations within the bet window — typically 10 to 15 seconds.
• The casino must not detect the pattern in the player’s betting and shut them out.

Tosa’s reported approach satisfies all four, which is itself the remarkable claim. For a deeper dive into the underlying probability mathematics, see our technical guide to roulette mathematics.

The Ritz Casino Case

In March 2004, Tosa and two associates — a Hungarian woman and a Serbian man — won approximately £1.3 million across two nights of play at the Ritz Casino in London. The casino noticed unusual betting patterns, reviewed the surveillance footage, and called the police. The three were arrested under suspicion of cheating.

The case became famous when, nine months later, the Crown Prosecution Service dropped all charges. The team had used no devices, no inside help, no tampering. They had simply watched the wheel and bet on sectors. Under UK gambling law at the time, that did not constitute cheating — the casino had taken bets on a wheel it presented as fair, and the players had assessed it more accurately than the casino did. The winnings were returned. The full story sits in our Ritz Casino feature.

The case did not change UK law immediately, but it did change casino operations across Europe. Wheel rotation schedules tightened. Bet-window cutoffs moved earlier in the spin to limit the observation time. Surveillance of betting patterns became more sophisticated. The window Tosa exploited has narrowed considerably since 2004 — though, by most accounts, it has not closed entirely on every wheel.

Why This Does Not Work Online

This needs stating clearly because it’s the most common misreading of Tosa’s work. Sector prediction applies to physical wheels in physical casinos. It does not apply to online RNG roulette, and it applies in only the most marginal sense to live dealer roulette streamed from a studio.

On RNG roulette, there is no physical wheel. The result is generated by a certified random number generator before the animation plays. There are no initial conditions to measure, no friction, no drop point — just a number selected by software that is independently audited for uniformity. No amount of observation tells you anything about the next result, because the next result has not yet been generated and the generator does not depend on observable inputs.

On live dealer roulette, there is a real wheel, but several factors close the advantage-play window:

• Camera angles. The viewer sees the wheel only from the angles the studio chooses, often with the rotor partially obscured.
• Latency. Online streams arrive with several seconds of delay. By the time you’ve assessed the spin on screen, the bet window has closed.
• Bet timer. Live studios cut bets earlier in the spin than physical casinos do, specifically to prevent visual ballistics.
• Wheel rotation. Studios rotate physical wheels frequently and run automated bias detection on every wheel they operate.

For the practical implications on different formats, see our live dealer roulette guide.

Legacy and the Advantage-Play Tradition

Tosa belongs to a small, persistent tradition of players who treat roulette as a measurement problem. Joseph Jagger broke the Bank at Monte Carlo in 1873 by identifying a biased wheel through six clerks recording spins for him. The García-Pelayo family in 1990s Madrid built a similar approach into a multi-year operation across European casinos. The Eudaemons — a group of Santa Cruz physics graduate students — wore the first computers ever taken into a casino in the 1970s, predicting drop sectors in real time. Tosa’s work sits squarely in this line.

The lesson from all of them is the same. Roulette is not unbeatable. It is, however, only beatable under conditions that almost no recreational player has access to: an unmaintained physical wheel, a long observation window, a casino that doesn’t pay attention, and a player with the mathematical training to convert observation into bets in real time. Strip away any one of those conditions and the edge disappears. That’s why these stories are so rare, and why the regular advice on this site — choose European over American, understand the payouts, manage your bankroll, treat any “winning system” with skepticism — is the right advice for the kind of roulette nearly everyone actually plays.

For more on the boundary between legitimate advantage play and outright cheating, our guide on wheel manipulation and our wheel bias & dealer signature feature cover the technical detail.

Frequently Asked Questions

Did Niko Tosa cheat?

No. The UK Crown Prosecution Service investigated his Ritz Casino winnings, found no evidence of cheating, dropped all charges, and ordered the casino to return the £1.3 million. He used no devices, no inside help, no wheel tampering. He observed the physical wheel and bet on sectors based on that observation. Under UK gambling law, that is not cheating — it is informed play on equipment the casino chose to present as fair.

Can I learn his method?

The technique — visual ballistics and sector prediction — is publicly documented and has been studied in peer-reviewed physics papers. Learning the underlying physics is not particularly difficult. Applying it profitably is. It requires hundreds of hours of observation per wheel, real-time calculation under time pressure, and access to physical wheels in casinos that haven’t tightened their bet windows. The skill ceiling is high, the legal venues are few, and modern surveillance makes sustained play increasingly difficult. Most players who attempt it without serious preparation produce no better results than random betting.

Does his method work on online roulette?

No. RNG online roulette uses a random number generator with no observable physical state — there is nothing to measure. Live dealer roulette uses real wheels but with camera angles, broadcast latency, and earlier bet cutoffs that close the observation window. The method is fundamentally tied to direct, real-time access to a physical wheel under non-rushed conditions, which online play does not provide.

How is sector prediction different from hot and cold numbers?

Completely different mechanisms. Hot and cold number tracking looks at past results to infer future ones, which is the gambler’s fallacy — past spins do not affect future spins on a fair wheel. Sector prediction ignores history entirely and measures the current spin’s physical state to predict where the ball will land in that spin. The two methods agree on nothing. One is a misreading of probability; the other is applied classical mechanics.

Are there other modern advantage players like him?

A small number, but most stay anonymous because publicity ends their careers — casinos share player photos across operators. The historical names that are public include Gonzalo García-Pelayo, the Eudaemons group (Doyne Farmer and colleagues), and the various teams reported in academic physics literature. The 2012 paper by Small and Tse formalised the underlying math. There is almost certainly ongoing modern advantage play that never reaches the news because the players prefer it that way.

What can recreational players actually take from his work?

Three things. First, roulette can be beaten, but only under conditions you almost certainly don’t have. Second, every claim of a “winning system” not grounded in physical observation of a specific wheel is mathematical fiction — see our coverage of Martingale, Fibonacci, and the rest of the strategies hub. Third, the casinos already know all of this, which is why physical wheels are rotated, audited, and surrounded by bet-window timing rules. The edge Tosa exploited still exists in principle but has gotten smaller, harder to find, and faster to disappear once spotted.