The Most Drawn Number in Every Lottery — and Why That Question Has No Fixed Answer

"What is the most drawn number in the lottery?" is one of the most searched lottery questions in the world, and almost every page answering it makes the same mistake: it prints a number, in bold, as though the number were a property of the game.

It is not. It is a property of the window you looked at, and it changes.

This site publishes frequency statistics computed from imported real draw history, game by game, on the live pages linked below. What it will not do is print a static "the most drawn number is 23" here, because by the time you read it the sentence may be false, and because — as the arithmetic below shows — the answer carries no information anyway.

What the question is really asking

Strip away the folklore and the question is: across N balls, which one has the largest count in the draw history?

Someone always has the largest count. That is not evidence of anything. In any set of numbers, one is the biggest. The only interesting question is how big a lead you should expect from a perfectly fair machine — because if the observed leader's margin is inside that range, the leader is noise.

That question has a precise answer, and it comes from order statistics.

The arithmetic: the maximum of N binomial counts

Consider a game drawing k balls from a pool of N, over D draws. For any individual ball:

• Probability of appearing in a given draw: p = k / N
• Its count over D draws is approximately Binomial(D, p)
• Mean count: μ = D × p
• Standard deviation: σ = √(D × p × (1 − p))

The expected maximum across all N balls is well approximated by Blom's formula for the expected largest of N normal variates:

E[max] ≈ μ + σ × z, where z = Φ⁻¹((N − 0.375) / (N + 0.25))

Worked example: UK Lotto (6 from 59), 1,000 draws

Roughly ten years of twice-weekly draws.

• p = 6 / 59 = 0.101695
• μ = 1000 × 0.101695 = 101.7 appearances
• σ = √(1000 × 0.101695 × 0.898305) = √91.35 = 9.56
• N = 59, so z = Φ⁻¹(58.625 / 59.25) = Φ⁻¹(0.98945) = 2.31
• E[max] ≈ 101.7 + 2.31 × 9.56 = 123.8
• E[min] ≈ 101.7 − 2.31 × 9.56 = 79.6

So from a flawless machine you should expect a hottest ball on about 124 appearances against an average of 102 — a lead of roughly 22% — and a coldest ball on about 80. The hottest number appears about 55% more often than the coldest, and nothing whatsoever is wrong.

Worked example: Powerball white balls (5 from 69), 1,000 draws

• p = 5 / 69 = 0.072464
• μ = 72.5, σ = √67.22 = 8.20
• N = 69, so z = Φ⁻¹(68.625 / 69.25) = 2.36
• E[max] ≈ 72.5 + 19.4 = 91.9
• E[min] ≈ 72.5 − 19.4 = 53.1

The leader runs 27% above average and appears 73% more often than the laggard. Again: a fair machine, correctly modelled, produces exactly this.

The lead shrinks — slowly

The mean grows with D. The spread grows with √D. So the leader's relative advantage decays, but at a pace that will outlast you. For UK Lotto's 6-from-59 matrix:

Ten thousand twice-weekly draws is about a century, and even then the expected leader is still 7% clear. There is no draw count at which a fair lottery produces a flat frequency chart. Anyone presenting an uneven chart as evidence of bias is presenting the expected output of fairness.

Which ball leads is uniform

One more result kills the strategy version of the question. By symmetry, every ball in a fair game has probability 1 / N of being the current leader — about 1.7% for a 59-ball pool. Leadership rotates. A ball that led after 500 draws has no better claim to leading after 1,000 than any other, and the count it accumulated is already in the past: it does not affect the next draw, which is the whole content of the gambler's fallacy.

The test that actually answers "is this machine fair?"

Comparing the leader to the mean is the wrong instrument, because it silently performs N comparisons and reports only the most extreme. The right instrument aggregates all N counts into one statistic: the chi-square goodness-of-fit test, whose expected value under a fair game is N − 1 (58 for UK Lotto, 68 for the Powerball white pool). The full method, including how to read the p-value without fooling yourself, is in testing hot numbers with chi-square. The mirror-image error — hunting for suspicious pairs across thousands of combinations — is in clustered pairs and multiple comparisons.

Physical causes of genuine bias do exist and are testable: ball weight, machine wear, air-jet pressure. How operators check for them is in ball weight and machine bias testing and physical draws vs software draws.

How to read our live pages

Frequencies on this site are computed from imported draw history, and the history is imported per game — so coverage differs by game and the window matters more than the leader.

• Powerball frequency statistics
• UK Lotto frequency statistics
• Frequency analyser — run the counts yourself and compare them against the expected spread above

When you look at one of those charts, do three things. Check the draw count, because a 200-draw window is nearly all noise. Compute μ and σ from the formulas above and see whether the leader sits inside μ + 2.3σ; if it does, it is exactly what a fair machine produces. Check whether the leader changed since the last time you looked — it usually has.

The honest answer

Every lottery has a most-drawn number. It is a fact about the past, it is expected to sit roughly 20–30% above average under perfect fairness, its identity rotates, and it has no bearing on the next draw. If a site tells you otherwise without showing the expected spread, it has skipped the only step that would have made the number mean something.

And if you were going to pick it anyway: number choice does not change your odds, but it does change how many people you split with — see is 1-2-3-4-5-6 less likely? and the number sharing risk tool.