Coin Flip FAQ
People ask the same coin-flip questions in two different voices. One voice wants permission to ignore a result. The other wants the probability. This page answers the probability. How to define a yes/no choice and then stop is covered in online coin flip for yes/no decisions. How a coin compares with dice and number ranges is covered in coin flip vs dice vs RNG.
The numbers below that are written as fractions are ideal fair-coin arithmetic: each flip Heads or Tails with probability 1/2, independently. They are not a survey of this website’s users, and they are not a claim about your pocket change unless a cited study says so.
Is an online coin flip the same as tossing metal?
No. The coin flip page animates a coin and chooses the face in the browser. A single flip calls Math.random() and maps a value below 0.5 to Heads and any other value to Tails. MDN documents Math.random() as a pseudo-random number with approximately uniform distribution from 0 inclusive to 1 exclusive, and it warns that the function is not cryptographically secure. Session streak counts live in the tab. A refresh resets them. The site does not keep a history of your flips.
A draw from 1 to 2 is the same two-label idea if you would rather see digits than a coin. Use the password generator or the UUID generator when the output is a secret or an identifier, not a dinner plan.
What does a fair coin mean in probability?
A fair coin model has two properties.
Equal weight. P(Heads) = P(Tails) = 1/2.
Independence. The probability of Heads on the next flip is 1/2 no matter what the previous flips were. The coin does not owe the room a Tail.
From those two statements you can compute the rest. In n flips, the number of Heads follows a binomial distribution. The probability of exactly k Heads is C(n, k) / 2n, where C(n, k) is the number of ways to choose k positions out of n. The expected number of Heads in 100 flips is 50. The standard deviation is the square root of n × (1/2) × (1/2), which for 100 flips is 5. A result of 55 Heads in 100 flips is one standard deviation above the mean. It is ordinary, not evidence that the coin changed.
Why do streaks feel wrong?
A run of Heads looks non-random because people expect a short sequence to contain a miniature copy of the long-run 50/50. Tversky and Kahneman called a related habit “belief in the law of small numbers” (Psychological Bulletin, 1971): people treat small samples as highly representative of the population the samples come from. The DOI is 10.1037/h0031322. That paper is about judgment, not a measurement of this coin tool.
The ideal math of a specific streak is simple. The probability that the next k flips are all Heads is 1/2k.
| Streak of Heads, specified in advance | Probability |
|---|---|
| 5 in a row | 1/32 = 0.03125 |
| 7 in a row | 1/128 = 0.0078125 |
| 10 in a row | 1/1024 ≈ 0.0009766 |
Those probabilities apply to a streak you specified before looking (“the next ten flips”). If you flip for a long time and then point at a streak you noticed, you had many windows in which a streak could start. Noticing one is less surprising than predicting that exact window. The next flip is still 1/2.
The chance that n flips are not all the same face is 1 − 2/2n. In 7 flips, the chance of seeing both faces at least once is 1 − 2/128 = 126/128 = 0.984375. The chance the whole seven are Heads or the whole seven are Tails is 1/64.
Does best-of-3 or best-of-5 change the odds?
On this site, best-of-3 flips three times and best-of-5 flips five times. The face that appears more often wins. Because the count is odd, there is no tie.
Heads wins a fair best-of-3 when at least two of the three flips are Heads:
- Exactly two Heads: C(3, 2) / 8 = 3/8.
- Three Heads: 1/8.
- Total: 1/2.
Heads wins a fair best-of-5 when at least three of five are Heads: (C(5, 3) + C(5, 4) + C(5, 5)) / 32 = (10 + 5 + 1) / 32 = 16/32 = 1/2.
A longer series still picks a winner at 1/2. It spends more flips to do it. In three flips, the probability of a 3–0 result in either direction is 1/4, and the probability of a 2–1 result in either direction is 3/4. People often trust the split more. The probability that Heads is the winner did not move.
For opening a match, write what the series controls and what the later games do. That script is how to run a best-of-series coin flip. The series math above is not a sport’s rulebook.
Are real coins biased?
A physics model says a vigorously flipped coin that is caught in the hand is slightly more likely to land on the face that was up at the start. Diaconis, Holmes, and Montgomery reported that chance as about 0.51 for natural flips (SIAM Review, 2007): doi.org/10.1137/S0036144504446436. Their analysis is about the motion of the toss, including wobble, not about a worn coin that is spun on a table.
Bartoš and colleagues tested that prediction with 350,757 recorded flips. They reported Pr(same side) = 0.508, with a 95% credible interval of 0.506 to 0.509. They also reported Pr(Heads) = 0.500 when the starting side was randomized, with a 95% credible interval of 0.498 to 0.502. The preprint is arXiv:2310.04153. They describe variation between people, and a same-side effect that shrank as flippers practiced. Those figures describe the coins and the catching protocol in that project. They do not measure Math.random(), and they do not say every pocket coin is 50.8% same-side.
If you need a physical toss to be as close as practical to a heads/tails balance, randomize the starting face (or ignore it by alternating who sets the coin) and catch the coin under a rule you stated first. If a one-percentage-point same-side effect would matter, do not use an unrecorded pocket toss for that decision.
Can the next flip be predicted from the streak counter?
The streak counter describes the current browser session. It is not an input to the next draw. The next single flip calls Math.random() again. Past Heads on the screen do not enter that call.
Math.random() is still a deterministic algorithm inside the browser, seeded by the engine, which is why MDN tells you not to use it for security. Predicting it is a different topic from the gambler’s fallacy, and it is not something this page tries to enable. For a yes/no among friends, the practical protection is shared observation: everyone sees the same click and you do not click until the preferred face appears.
What should I use instead of a coin?
A coin has two outcomes. The moment you have a third, the coin is the wrong shape.
- Several named options: wheel spinner.
- People: name picker, with unique picks if one click must not repeat a name.
- Dice notation such as 2d6: dice roller.
- Any integer range: the random number generator.
A raffle with more than two entrants is a list problem. Freeze the names, draw once, and keep the screenshot. The steps are in how to run a fair online raffle. Two entrants can use a coin if both agree. Three entrants cannot, unless you are willing to invent an unequal procedure such as “Heads wins for person A, Tails flips again for B versus C,” which gives A a 1/2 chance and B and C a 1/4 chance each. Say that inequality out loud or, better, put all three names in the picker so each has 1/3.
A short checklist before you trust a result
- There are exactly two acceptable outcomes, already written down.
- You will flip once on a method everyone can see.
- A rerun rule exists only for a failed toss or a failed page, not for a disliked face.
- If the stake is a prize or a record, you are following a published rule, not improvising.
- If you needed three or more outcomes, you switched tools before clicking.
References
- Diaconis, P., Holmes, S., & Montgomery, R. (2007). Dynamical bias in the coin toss. SIAM Review. https://doi.org/10.1137/S0036144504446436
- Bartoš, F., et al. Fair coins tend to land on the same side they started: Evidence from 350,757 flips. arXiv:2310.04153
- Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychological Bulletin, 76(2), 105–110. https://doi.org/10.1037/h0031322
- Math.random() — MDN
Disclaimer: Probability statements labeled as ideal fair-coin math assume independent flips with P(Heads) = 1/2. Cited studies describe their own coins and protocols. This page is not legal, gambling, or security advice.
FAQ
Is a fair coin exactly 50/50?
In the ideal model, yes: Heads and Tails each have probability 1/2, and flips are independent. A caught physical coin can show a small same-side bias. This site’s online coin is a browser draw, not that physical experiment.
Does a streak of heads make tails due?
No. Under an ideal fair coin, the next flip is still 1/2 after any history. A streak changes how the past looks. It does not change the probability of the next face.
How rare is ten heads in a row?
On an ideal fair coin, ten specific flips all landing Heads has probability 1/1024, about 0.098%. In a long session, some streak of that length becomes more plausible because you have many chances to start one.
Can I use a coin flip to pick a raffle winner?
Only when there are exactly two entrants and both accept the method. More names need a name picker or a numbered draw with a published list, not a two-face coin.