On a fair wheel no number hits most. We simulated hot and cold numbers from 37 to 37,000 spins, tested whether they predict anything, and show how to check a wheel for bias.
On a fair wheel no number hits most. We simulated hot and cold numbers from 37 to 37,000 spins, tested whether they predict anything, and show how to check a wheel for bias.
No number hits most on a fair roulette wheel: each of the 37 numbers has the same 1 in 37 chance on every spin, and hot and cold numbers are simply what that randomness looks like over a short window. In our simulation, the hottest number of 370 spins came up 17.3 times on average against 10 expected, and over the next 370 spins it hit 9.99 times, exactly like any other number.
This guide shows how far numbers drift from the average over 37, 370, 3,700 and 37,000 spins, what the gambler’s fallacy looks like in 10 million spins, when a wheel really can favor a number, how to read the hot and cold panel of a live table, and a simple fairness check you can run yourself. Every simulated figure comes from a fair European wheel unless it is clearly labeled as a hypothetical biased one.
Contents
Over any window of spins, some number will always be on top, and some number will always be missing. That is not a sign that the wheel prefers it. With 37 numbers sharing a few hundred spins, the counts are bound to spread out, and the leader changes as more spins come in.
The only way a number can truly hit more often is a physical fault in the wheel: a pocket, a fret or a tilt that nudges the ball. Those wheels have existed and players have profited from them, but finding one takes tens of thousands of recorded spins, and modern regulated wheels are built and checked to prevent it. Everything else, from “hot streaks” to numbers that are “due”, is the normal noise of a fair wheel.
We simulated 10,000 tables for each session length and recorded the hottest and the coldest number of each table. The expected count is the number of spins divided by 37; the standard deviation tells you how far a typical number strays from it.
| Spins | Expected hits per number | Typical spread (1 standard deviation) | Average hottest number | Average coldest number |
|---|---|---|---|---|
| 37 | 1 | 1.0 | 3.6 hits | 0 hits (13.4 numbers never appear) |
| 370 | 10 | 3.1 | 17.3 hits (+73%) | 4.0 hits (-60%) |
| 3,700 | 100 | 9.9 | 121.9 hits (+22%) | 79.4 hits (-21%) |
| 37,000 | 1,000 | 31.2 | 1,067.9 hits (+7%) | 933.4 hits (-7%) |
Two things happen as spins add up. The gap in hits grows: the hottest number leads the average by about 7 hits after 370 spins and by about 68 after 37,000. But the gap as a share of the expected count shrinks, from +73% to +7%. Short samples exaggerate differences; long samples flatten them. After 37 spins, more than a third of the numbers have not appeared at all, so a “cold” number at that point means nothing.
The extremes go further than the averages. Across the 10,000 tables of 370 spins, the hottest number reached 27 hits at most, and some tables had a number with no hits at all. Across the tables of 37,000 spins, counts ranged from 867 to 1,152, all from a perfectly fair wheel.
Here is one of those tables, a single run of 370 spins. Number 16 came up 18 times, number 6 came up 17 times, and number 24 only 4 times. On a live screen, 16 would be flagged as the hottest number and 24 as the coldest.
A fairness test on these counts (explained below) gives a chi-square of 39.4, well under the 51.0 alarm level: this spread is exactly what a fair wheel produces. Nothing about 16 or 24 is special. Run another 370 spins and different numbers take their places.
The real question is whether a hot or cold number tells you anything about the next spins. We ran 100,000 tables: 370 spins to find the hottest and the coldest number, then 370 more spins to see what they did next. Number 17 was tracked as a control.
| Number | Hits in the first 370 spins | Hits in the next 370 spins | Hit rate next |
|---|---|---|---|
| Hottest of the first 370 | 17.31 | 9.99 | 2.70% |
| Coldest of the first 370 | 3.98 | 10.01 | 2.71% |
| Number 17 (control) | 10.00 expected | 9.99 | 2.70% |
Hot numbers cooled down to exactly average, and cold numbers warmed up to exactly average. Neither kept its streak and neither “caught up”. Over the following 37 spins each of them hit 1.00 times on average, the same 1 in 37 as any number. On a fair wheel, hot and cold labels describe the past and carry no information about the next spin.
The gambler’s fallacy is the belief that a result becomes more likely because it has not happened for a while, or less likely because it just happened several times. It is also called the Monte Carlo fallacy, after a famous night at the Monte Carlo Casino on August 18, 1913, when the ball landed on black 26 times in a row and players lost heavily betting that red was due. The chance that 26 given spins are all black is (18/37)26, about 1 in 136.8 million.
We ran 10,000,000 spins on a fair wheel and looked at what followed streaks and long absences.
| Situation | Cases in 10 million spins | Next spin |
|---|---|---|
| After 5 reds in a row | 139,457 | red 48.6%, black 48.7%, zero 2.7% |
| After 8 reds in a row | 15,773 | red 48.3%, black 48.8%, zero 2.8% |
| After 10 reds in a row | 3,704 | red 49.8%, black 47.5%, zero 2.6% |
| A number absent for 100+ spins | 23,953,829 | hit 2.69% |
| A number absent for 200+ spins | 1,564,554 | hit 2.68% |
The fair-wheel values are 48.65% red, 48.65% black, 2.70% zero, and 2.70% for any single number. Every line matches them within the noise you expect from the number of cases. A streak of 21 reds also appeared in these 10 million spins: long runs are rare on any given spin but certain to show up somewhere when you watch long enough. The roulette streak calculator gives the odds for any streak and session length.
Biased wheels are real history. In 1873 the English engineer Joseph Jagger is reported to have recorded the results of the wheels at Monte Carlo and bet on the numbers one wheel favored. In the early 1990s Gonzalo Garcia-Pelayo used computers to analyze large numbers of spins at the Casino de Madrid; the casino took him to court and lost. In 1986 a team led by Billy Walters won $3.8 million at the Golden Nugget in Atlantic City, and casinos moved to low-profile wheels afterward.
Today’s regulated live studios work under audit rules. The UK Gambling Commission, for example, requires that “live dealer operations must be fair and independently auditable”, asks for equipment of commercial casino quality with staff responsible for monitoring its integrity, and asks that game logs be collated into statistics so trends can be analysed. Its guidance also notes the value of measuring the distribution of results on a roulette wheel over an extended period. A studio that logs every spin has far more data to spot a fault than a player watching a few hundred results.
How long would it take you to spot a bias? We built a hypothetical faulty wheel where pocket 17 gets extra weight and the other 36 pockets share the rest, and checked how often a fairness test caught it (2,000 runs per cell).
| Hypothetical bias on 17 | Return of a bet on 17 | 370 spins | 3,700 spins | 37,000 spins |
|---|---|---|---|---|
| 10% more often | +7.0% | caught 5.5% | caught 6.3% | caught 29.5% |
| 25% more often | +21.6% | caught 6.7% | caught 19.6% | caught 100% |
| 50% more often | +45.9% | caught 10.5% | caught 76.4% | caught 100% |
Even a fault big enough to give the player a large edge is nearly invisible after 370 spins: the test fires only 5.5% to 10.5% of the time, close to its 5% false-alarm rate on a fair wheel. Only a gross 50% fault shows up reliably by 3,700 spins, and a subtle 10% fault is still missed most of the time after 37,000. A hot number on a screen with a few hundred spins behind it is not evidence of anything.
Most live roulette games show recent results and a statistics panel with hot and cold numbers over a recent window of spins; the window size varies by game. Our guide to live roulette games covers the main variants. A few rules of thumb from the simulations above:
If you want to keep your own counts, the free roulette number tracker shows hot, cold, dozens and columns as you enter spins, and our live roulette statistics page explains the figures most tables display.
The chi-square test answers one question: are these counts more uneven than a fair wheel usually produces? You can do it with a spreadsheet.
A worked example from one simulated run of 3,700 spins (E = 100): number 12 hit 125 times, which adds (125 – 100)2 / 100 = 6.25. Number 5 hit 77 times and adds 5.29, number 9 hit 80 times and adds 4.00. All 37 numbers together give 44.8, below 51.0, so the wheel passes, as it should, because it is fair.
Two cautions. A fair wheel fails the 5% test 5% of the time by design: in our simulation 4.7% to 5.2% of fair tables were flagged, at every length. And as the biased-wheel table shows, the test needs thousands of spins to see anything subtle. The limits of prediction in general are covered in can AI predict roulette.
RG2030 keeps these counts for the live roulette you play, in your browser on any device. From the Pro Pack its statistics include dozens, columns and hot and cold numbers on your table, plus Interesting Facts that show what followed the last number there, and the statistics start after 500 numbers of that table.
Read them the way this guide suggests: as an exact record of your table, kept without effort, not as a forecast. On a fair wheel the house edge on the next spin stays at 2.70% whatever the panel says. The Packs are listed on the pricing page.
RG2030 collects the numbers of the live roulette you are playing and, after 500 numbers, shows statistics and suggestions built on that table, only when a bet reaches the Win Rate you choose. It runs in your browser on any device; the optional RG2030 Bridge addon for Chrome on a computer reads the numbers automatically and enables Autobet. You pay $1 per minute, with no subscription. It cannot promise wins: it gives you data to bet with more information.
On a fair wheel, none: every number has a 1 in 37 chance (European) or 1 in 38 (American) on every spin. Over a short window one number always leads, but in our simulation the hottest number of 370 spins hit at the normal 2.70% rate over the next 370.
No. Across 100,000 simulated tables, the hottest number of 370 spins averaged 17.31 hits, then 9.99 hits in the next 370 spins, the same as any number. Hot streaks describe the past only.
No. A number absent for 100 spins still has a 2.70% chance on the next spin. In our 10-million-spin simulation, numbers absent for 100 or more spins hit the next spin 2.69% of the time.
Thousands to tens of thousands. In our test, a hypothetical wheel favoring one number by 50% was caught 76.4% of the time after 3,700 spins, and a 10% bias was caught only 29.5% of the time after 37,000 spins.
Regulated live studios are audited. The UK Gambling Commission requires live dealer operations to be fair and independently auditable, with staff monitoring the equipment and game logs collated into statistics, and it notes the value of measuring the distribution of results on a roulette wheel over time.
Roulette is a game of chance for adults (18+): set a limit before you play, and visit our responsible gambling page if you need help.