Watch a roulette table for ten minutes and you will see it: the same number twice in a few spins, one part of the wheel getting hit again and again, other numbers that never seem to show up. It is the idea behind "hot numbers", "cluster theory" and the apps that track which section of the wheel a dealer hits. If the numbers bunch up, surely you can bet on the bunch?
We tested it. The short answer: the clusters are completely real — and completely useless for winning.
Part 1: Yes, numbers cluster
Most people expect 37 numbers to spread out fairly evenly. They don't, and it is pure maths. It is the same effect as the famous birthday problem: in a room of just 23 people, two share a birthday more often than not. On a European wheel the chance that some number repeats grows fast:
| Spins | Chance at least one number repeats |
|---|---|
| 5 | 24.6% |
| 8 | 55.7% |
| 10 | 73.7% |
| 12 | 86.5% |
After only eight spins, a repeat is more likely than not. And over 37 spins — one "full round" of the wheel — you will on average see only 23.6 different numbers. Around 13 numbers don't show up at all, while others come two or three times. That uneven pattern is what a random wheel looks like. An evenly spread wheel would be the suspicious one.
Part 2: But betting on them doesn't work
Seeing clusters afterwards is easy. The question is whether you can use them before the next spin. We let our engine play 20 million spins on a European wheel and tested six ways of picking what to bet on — always $1 on the chosen number (or $1 spread over a 9-number section of the wheel):
| How the bet was chosen | Hit rate | Expected | Return per $1 |
|---|---|---|---|
| Hot number (most hits in the last 37 spins) | 2.71% | 2.70% | −2.6% |
| Cold number (fewest hits in the last 37) | 2.71% | 2.70% | −2.5% |
| Repeat the last number | 2.70% | 2.70% | −2.7% |
| A random number | 2.70% | 2.70% | −2.7% |
| Section of 9 around the last number | 24.34% | 24.32% | −2.6% |
| A random section of 9 | 24.33% | 24.32% | −2.7% |
Every method lands on the textbook number. Hot numbers don't stay hot, cold numbers aren't "due", the last number is no more likely to repeat than any other, and the section around the last number is hit exactly as often as any random section. The tiny differences in the last column are simulation noise — run it again and they shuffle around. Every way of choosing loses the same 2.7% house edge.
Why the clusters don't help
A roulette wheel has no memory. The ball doesn't know which numbers came up before, so the chance for every number on the next spin is exactly 1 in 37 — no matter what the last 37 spins looked like. Clusters only exist when you look backwards. They tell you that randomness is lumpy, not where the next lump will be.
That is also why "a repeat is likely within 8 spins" doesn't help: you know some number will probably repeat, but not which one. To cash in, you would have to bet on all of them — and then you are just paying the house edge on every number.
What about dealer signatures and wheel bias?
There is one real exception in history: physically faulty wheels. Decades ago, a few players found worn or tilted wheels that favoured certain numbers, and won by recording thousands of spins. Modern casinos check and rotate their wheels, and online and automated wheels are tested and certified for exactly this.
Apps that record a dealer's last few dozen spins and point to a "hot section" run into the problem from Part 1: in a small sample, random clusters always appear. Spotting a genuine bias would take thousands of spins on the same wheel — and even then the edge has to be big enough to beat 2.7%. A section that looks hot over 30 spins is, almost always, just what randomness looks like.
Our verdict
Numbers do cluster, and now you know why: it is the birthday problem at work, not a pattern. Betting on hot numbers, cold numbers, repeats or sections changes nothing — every choice hits at the same rate and loses the same 2.7%. Pick your numbers however you like; it is part of the fun. Just don't pay for a theory, a book or an app that promises the clusters will tell you where the ball goes next. If you want to see what a system really does, test it in the System Builder or browse the 100 Club — every one of them is measured the same way.