Every roulette scoreboard shows "hot" and "cold" numbers, and many players read them as a forecast. This guide uses real arithmetic to show how much numbers drift apart by chance, what a genuine anomaly would look like, and how many spins it takes to tell the two apart.

Can you predict roulette numbers?

On a fair wheel, no. Each spin lands on one of 37 pockets on a European wheel (38 on an American one), and the ball has no memory of where it landed before.

Prediction systems sell the opposite idea. They track recent numbers, find the ones that "keep coming up" or the ones "due" to appear, and tell you to bet accordingly. Neither version changes the odds. A straight-up bet wins with probability 1/37 on every spin, pays 35 to 1, and loses 2.70% of each stake on average. The roulette rules and payouts page lists every bet, and all of them carry that same edge on a European wheel.

We cover the wider question in can roulette be predicted?. This post focuses on number data: what a frequency chart can tell you and what it can't.

What are hot and cold numbers in roulette?

A hot number is one that has landed more often than average over a recent window. A cold number has landed less often, or not at all.

The labels describe the past. They make no claim about the next spin, even though the display invites you to read them that way. If the wheel is fair, a number that hit 18 times in the last 370 spins has the same 1/37 chance on spin 371 as a number that hit 4 times.

Belief in cold numbers being "due" is the gambler's fallacy. Belief in hot numbers "staying hot" is its mirror image. Both treat random clustering as a signal. Our post on why winning streaks can fool you explains why the brain falls for both.

How much do roulette numbers vary by chance?

Take 370 spins on a fair European wheel, roughly several hours at one live table. Each number should appear 10 times on average (370 ÷ 37). The standard deviation for a single number is about 3.1 hits, so a lot of spread is normal.

Here is what a fair wheel produces across all 37 numbers in 370 spins, based on exact binomial odds and a 40,000-run simulation:

What you see after 370 spins	Expected on a fair wheel
Average hits per number	10
Most-hit number (typical result)	17 hits
Chance at least one number hits 16+ times	87%
Chance at least one number hits 20+ times	11%
Least-hit number (typical result)	4 hits
Chance at least one number hits 4 or fewer times	67%
Numbers with 15+ hits (average count)	3.0
Numbers with 5 or fewer hits (average count)	2.4
Chance a given number lands between 7 and 13 hits	74%

So on a perfectly fair wheel, the top number usually lands about 70% more often than average, and the bottom number about 60% less often. Three "hot" numbers and two or three "cold" ones show up in a typical sample with no cause behind them.

Short windows look even more extreme. In 37 spins, the average number of pockets that never appear is about 13.4. A scoreboard showing a third of the wheel as "cold" after 37 spins is what fairness looks like.

Does wheel bias make some numbers hit more often?

It can, in principle. A worn bearing, a loose fret, or a tilted wheel head can push the ball toward some pockets. That is the one physical route to a real number edge, and we cover it in detail in can a roulette wheel really have a bias?

In practice, modern wheels make this rare. Casinos use low-profile pocket separators, balanced rotors, and level checks. Many track their own number frequencies and rotate or service wheels.

The key point for number data: a real bias and ordinary variance look identical in a small sample. The table above shows how loud chance alone can be.

What did Joseph Jagger and the Pelayos find?

Two well-documented cases show what a real bias took to find and exploit.

Joseph Jagger, 1873: An engineer from Yorkshire, Jagger hired clerks to record results at the Beaux-Arts casino in Monte Carlo. Their records showed one wheel favoring certain numbers, and he bet on them with success until the casino started moving wheels between tables and changing parts.
Gonzalo Garcia-Pelayo, early 1990s: Garcia-Pelayo and his family recorded thousands of spins at Casino Gran Madrid and used computer analysis to find wheels that favored some numbers. The casino banned them. They took the case to court in Spain and won, and casinos began rotating wheels and switching to designs that resist this kind of wear.

Both cases share three features. The bias came from a physical defect, the players collected spins by the thousand before betting, and the casinos closed the gap once they noticed. Nobody found an edge by glancing at a hot-numbers board.

Do roulette sectors and neighbors predict anything?

Some players group numbers by their position on the wheel, called sectors or neighbors, instead of by number. The logic is that a physical bias would affect a region of the wheel, not one isolated pocket.

That logic holds for a biased wheel. It does not rescue a fair one. A 5-pocket sector should catch about 50 of 370 spins, with a standard deviation near 6.6 hits. A sector that catches 60 spins looks "hot" but sits inside normal variation.

Sector betting also pays the same house edge as any other combination. Five straight-up bets on neighbors cost 2.70% of the total stake on average, the same as a bet on red. If you want a wider bet, the dozen strategy guide covers how dozens behave.

How would you test a real roulette number anomaly?

A proper test asks one question: is this spread larger than chance would produce? Three ideas help.

Chi-square test: Compare the observed count for each of the 37 numbers against the expected count. With 36 degrees of freedom, a fair wheel scores above 51.0 only 5% of the time and above 58.6 only 1% of the time. A score below that is consistent with a fair wheel.
Account for 37 chances: If you scan all 37 numbers for the most extreme one, you will always find a standout. A three-standard-deviation threshold on each number keeps the chance of a false alarm across the whole wheel near 5%.
Sample size: Suppose one number hits 1 time in 30 instead of 1 in 37. That would turn its straight-up bet into a 20% player edge, a large bias. Detecting it at three standard deviations takes about 6,000 spins on average, and about 9,750 spins to catch it 80% of the time. Smaller biases need far more.

Then the test must repeat. A pattern found in one batch of spins has to show up again in fresh spins recorded afterward. Without that second check, you are fitting noise.

How should you use roulette number data responsibly?

Number data is useful for one thing: checking whether a table behaves like a fair game. It cannot tell you where the next ball will land.

Read hot and cold as history: They describe the last window and nothing more.
Expect wide spread: A number hitting 17 times in 370 spins is the typical top result on a fair wheel.
Treat systems with suspicion: Any method built on recent numbers keeps the same 2.70% edge on a European wheel and 5.26% on an American one. See roulette odds and house edge for the full picture.
Watch the clock and the budget: Tracking numbers can stretch a session. Set a loss limit and a time limit before you start.

If gambling stops feeling like entertainment, free help is available from BeGambleAware and the National Council on Problem Gambling.

What Golden Zero shows

Golden Zero records every result from the live tables it watches, tracks streaks, and sends streak alerts on Telegram. The live game analytics hub shows recent results from those tables, so you can compare real outcomes with the fair-wheel spread above. It reports what happened. It does not forecast the next spin.

Key Takeaways

On a fair wheel, every number has a 1/37 chance on every spin, whatever it did before.
In 370 fair spins, the most-hit number lands about 17 times and the least-hit about 4 times, with no cause behind it.
Real number bias comes from physical defects, and modern wheels and casino checks make it rare.
Jagger and the Pelayos recorded spins by the thousand before betting, and casinos changed their wheels in response.
Testing a real anomaly takes a chi-square check, thousands of spins, and a repeat on fresh data.

Keep Reading

Golden Zero shows you the live numbers as they land, so watch the data with clear eyes and play only within limits you can afford to lose.