Roulette bias checker
Is a wheel biased? Run the z-test
Enter how many spins you've tracked and how many times one number hit, and see a proper statistical z-test, the same style of maths Joseph Jagger's clerks were working toward by hand in 1873, run properly this time.
This tests one number you've already flagged as unusual against pure chance, it can't tell you which number to watch in the first place, and it can't turn a small sample into a reliable answer.
150 hits on one number across 5,000 spins · 38 pockets
Sample too small to say much
Expected hits on a fair wheel: 131.6 (±11.3 standard deviation). Your number is +1.63 standard deviations from that expectation.
Under 10,000 spins, even a genuinely fair wheel routinely produces deviations this size. Real wheel-bias hunters historically tracked tens of thousands of spins before acting on a suspicion.
How do you test a roulette wheel for bias?
A bias test compares how many times a specific number hit against how many times you'd expect it to hit on a fair wheel, given the number of spins you've tracked. The gap between those two figures, measured in standard deviations, is called a z-score. The bigger the z-score, the less likely the deviation happened purely by chance, and the calculator above works that number out directly from your spins and hits, exactly the statistic that real wheel-clocking has always been built on.
How many spins do you need to detect bias?
The calculator flags anything under 10,000 spins as "too small to say much", even when a number looks unusual, because small samples are naturally noisy: a perfectly fair wheel routinely throws up numbers that look 20% or 30% over expectation across a few hundred spins, purely from ordinary variance. The roulette AI hub page covers the real history here: Joseph Jagger's clerks tracked thousands of spins across multiple wheels at a Monte Carlo casino in 1873 before he felt confident enough to bet heavily on what they'd found, and even then it was a real manufacturing fault, not a hunch confirmed by a handful of results.
The trap almost everyone falls into
This is where casual "hot number" tracking usually goes wrong. A wheel has 37 or 38 numbers on it. Test every single one at a 95% confidence threshold and, purely by the maths of chance, roughly one number in twenty will cross that threshold on a completely fair wheel, every single time you run the test. Spotting one number that looks hot after the fact, then testing only that number, skips past this entirely and makes ordinary noise look like a discovery. Proper bias analysis checks the full spread of all 37 or 38 numbers together, not just the one that already caught your eye.
We saw this happen in our own data. In a million spins of a simulated wheel we know to be perfectly fair, number 6 came up 27,420 times, 2.42 standard deviations above expectation. Test 6 on its own and you'd call it biased with 98% confidence. Test all 37 numbers together, as a chi-square does, and the wheel scores 28.3 on 36 degrees of freedom: textbook fair. One "significant" number out of 37 is exactly what chance predicts.
Where wheel bias still matters
None of this applies to RNG-based online roulette, there's no physical wheel there to be biased in the first place, results come from a certified random number generator covered in the glossary's RNG entry. Live-dealer roulette does spin a real wheel, but modern wheels are precision-engineered and actively monitored, often with the same live dealer AI tracking covered on the hub page, specifically to catch and correct the kind of wear or manufacturing fault Jagger found by hand. How much weight you give a suspicious-looking streak is up to you; the maths above doesn't change either way.
Getting enough data
A proper bias test needs patience and a lot of data.
FAQ
Bias checking FAQ
How do you test a wheel for bias?+
By comparing a number's observed hit count against what a fair wheel would produce on average, using a standard statistical z-test. The calculator above runs that exact test, it's the same underlying method, if not the same scale, that Joseph Jagger's clerks were manually working toward in 1873.
How many spins do you need to detect bias?+
Tens of thousands, realistically. Historical wheel-clocking teams tracked upwards of 10,000 to 20,000 spins per wheel before acting on a suspected bias, because smaller samples produce large-looking deviations from pure chance constantly. A few hundred spins showing an unusual number proves nothing on its own.
What's the multiple comparisons problem with wheel bias?+
A wheel has 37 or 38 numbers, and testing each one at a 95% confidence threshold means roughly 1 in 20 numbers will look statistically "significant" purely by chance, even on a completely fair wheel. Spotting one unusual number after the fact and testing only that one is a well-known statistical trap, real bias hunting checks every number's deviation together, not just the one that already looked interesting.
Can online roulette have a biased wheel?+
No, RNG-based online roulette has no physical wheel to bias, results come from a certified random number generator. Live-dealer roulette does use a real spinning wheel, but modern wheels are precision engineered, regularly maintained and monitored specifically to prevent the kind of manufacturing fault Jagger exploited in 1873.
Is this tool the same as an AI roulette predictor?+
No, and that distinction matters. This is a standard statistical test applied to data you already have, it doesn't predict anything, it only checks whether a past pattern is more extreme than pure chance would produce. "AI roulette predictor" apps claim to forecast the next spin in advance on a fair wheel, which isn't possible and isn't what this calculator does.
Keep exploring
Related tools
More Roulette AI tools
Other games