Coin flip probability: streaks, runs and the gambler's fallacy

The odds of n heads in a row are (1/2)ⁿ, calculated before you start. But once you’ve already flipped nine heads, the next flip is still exactly 50/50, the coin has no memory. That single sentence is the whole answer, everything below is why both halves are true at once. Flip a coin, free, and watch it happen for yourself.

The basic maths

Each flip is an independent event, and the multiplication rule for independent events says the probability of a specific sequence is the product of each individual probability. For n heads in a row, that’s (1/2) × (1/2) × … n times, or (1/2)ⁿ:

Every row is exactly (1/2)ⁿ, and the “odds” column is exactly 2ⁿ, the two columns are the same number written two ways.

Notice how fast the probability collapses as the streak grows, doubling the streak length from 5 to 10 doesn’t double the rarity, it multiplies it by 32, from 1-in-32 to 1-in-1,024. That’s the nature of exponential decay, each additional flip in the streak halves the remaining probability, so long streaks become vanishingly rare far faster than intuition usually expects.

Flip a coin three times, every outcome

Three flips produce exactly eight equally likely sequences: HHH, HHT, HTH, HTT, THH, THT, TTH, TTT. Counting directly from that full list rather than a shortcut formula:

• Exactly two heads — HHT, HTH, THH, three of eight, 3/8, 37.5%.
• At least one head — every sequence except TTT, seven of eight, 7/8, 87.5%.
• All three the same — HHH or TTT, two of eight, 2/8, 25%.

The gambler’s fallacy

The coin has no memory. Every flip is independent, so nothing about the sequence so far changes the odds of the next one. The canonical illustration of getting this wrong: on 18 August 1913, the roulette wheel at the Monte Carlo Casino landed on black 26 times in a row. As the streak grew, gamblers piled increasingly large bets on red, certain it was “due” to balance things out, and lost heavily doing it, red didn’t appear until the 27th spin. On a European wheel (18 red, 18 black, 1 zero), the probability of either color landing 26 times running is 2 × (18/37)²⁶ ≈ 1 in 68.4 million, genuinely extraordinary, but every single spin was still independent of the ones before it. The event is documented as the origin of the term “Monte Carlo fallacy”, the more common name for exactly this mistake.

Before vs during, the distinction that fixes everything

Two different, equally true statements, and almost every confusing coin-flip question comes down to mixing them up:

• P(ten heads in a row) = 1/1,024, measured before the first of the ten flips happens. This is a statement about how rare that whole specific sequence is, viewed from the start.
• P(the tenth flip is heads, given the previous nine were heads) = 1/2, measured during, once nine of the ten flips have already happened. This is a statement about one single, independent flip, and the nine that came before it are irrelevant to it.

Both are correct. “What are the odds of flipping ten heads in a row?” is asking the first question. “I’ve flipped nine heads, what are the odds the next one is heads too?” is asking the second. They sound similar and the answers are wildly different, 1-in-1,024 versus 1-in-2, because they’re genuinely different questions.

The confusion almost always comes from swapping which question is actually being asked partway through a conversation. Someone points out that ten heads in a row is a 1-in-1,024 event, which is true, and someone else takes that to mean the eleventh flip is somehow more likely to be tails, to “balance out” that rarity, which isn’t true at all. The rarity belongs to the whole ten-flip sequence viewed as one event before it happened, not to any single flip within it, and once nine of those flips are already history, they’re no longer part of a probability calculation, they’re just facts.

The hot hand, and the opposite mistake

Believing a streak is “due” to break is the gambler’s fallacy. Believing a streak will keep going because it’s “on a roll” is the mirror-image mistake, sometimes called the hot-hand fallacy. For a fair coin, both are wrong for exactly the same reason, there is no mechanism by which past flips could influence the next one, so neither “it must end” nor “it must continue” has anything real behind it. The coin doesn’t know it’s on a streak any more than it knows it isn’t, there’s no physical mechanism, no memory and no momentum for either belief to attach to.

Best of three, and why it’s still 50/50

P(a given side wins a best-of-three match) = 50%, and separately, P(the match needs a third flip at all) = 50% too, because a 2–0 sweep and a 2–1 decision are equally likely outcomes. Working it out directly: a side wins 2–0 with probability 1/4, or wins 2–1 with probability 1/4 (two different orders, each 1/8), for a total win probability of 1/4 + 1/4 = 1/2. The match reaches a third flip exactly when neither side has won the first two, which also works out to 1/2. Go flip a coin and its running tally is built for exactly this, call your best-of-three live and watch the running heads-tails count settle the match.

Is a real coin 50/50?

Not quite, as the honest addendum. Physical coins carry a small, real same-side bias, Diaconis, Holmes and Montgomery’s 2007 physics model predicted it from the wobble in a tossed coin’s spin, and Bartoš et al.’s 2023 study of 350,757 real flips measured it empirically at about 50.8%. A digital flip drawn from a cryptographic random generator has no spin and no starting side, so there’s no physical mechanism for that particular bias to act through in the first place.

Teaching this with a coin and a die

A short classroom sequence that makes the maths concrete rather than abstract: have students predict how many heads they’ll get, flip 50 times and keep a running tally, then compare the actual result against the predicted spread. Repeat the same predict-then-test structure with two dice, the odds of any specific total when rolling two dice follow their own, different distribution, rolling a 7 is far likelier than rolling a 2 or a 12, which makes a genuinely useful contrast to a coin’s flat 50/50 once students have just spent ten minutes with the coin’s simpler case.

FAQ

What are the odds of 10 heads in a row?

(1/2)^10 = 1/1,024, about 0.0977%, measured before you start flipping. Once you’ve already flipped nine heads, the odds of the tenth also being heads are back to a plain 1/2, both numbers are correct, they just answer different questions.

Does a previous flip affect the next one?

No. Each flip is an independent event, a coin has no memory of what it landed on before. Nine heads in a row changes nothing about the odds of the tenth flip.

Is an online coin flip really random?

Ours draws from the browser’s cryptographic random number generator, a source built for genuine unpredictability rather than a simple pseudo-random formula. That’s an accurate, checkable claim, not a marketing one, “provably fair” or externally auditable is a different, stronger claim we don’t make.

What are the odds of getting two heads in three flips?

3/8, 37.5%, three of the eight equally likely outcomes (HHT, HTH, THH) have exactly two heads.

Is a real coin toss biased?

Slightly, yes. Physical coins carry a small, real bias toward landing on the side they started on, measured at about 50.8% across the largest study of real flips to date, caused by the physics of the toss itself. A digital flip has no toss to be biased in the first place.

Ready to test it yourself? Flip a coin, free, no signup, with a running tally that’s perfect for a best-of-three or a quick classroom demonstration. For a different distribution entirely, roll two dice and see how the odds spread out around 7 instead of sitting flat at 50/50, or use the random name picker for anything with more than two outcomes.