Is a Coin Flip Really 50/50? What 350,757 Flips Show

For everyday purposes, yes: a fair coin lands heads about half the time. But a coin flipped by hand and caught has a small, measurable tendency to land on the side it started on. Physics predicts about 51%, and an experiment with 350,757 flips measured 50.8%. An online coin has no starting side, so its odds come from the code: 50/50.

A coin toss is physics, not magic

Once a coin leaves your thumb, its path is fixed by how hard you flicked it, how fast it turns and how much it wobbles. It feels random only because nobody controls those starting conditions precisely enough to aim the result.

In 2007, Persi Diaconis, Susan Holmes and Richard Montgomery published “Dynamical Bias in the Coin Toss” in SIAM Review. Two points from that paper are worth knowing:

• A machine can make it predictable. They describe a spring-loaded coin-tossing machine that, carefully adjusted, made a coin started heads up land heads up every time.
• Human flips wobble. When a person flips a coin, its spin axis often drifts during flight (precession). In their model, any wobble makes the coin spend slightly more of its flight showing the face it started with. From high-speed film of real flips, they estimated that a naturally flipped coin caught in the hand comes up the way it started about 51% of the time.

That 51% rested on a small set of filmed flips (27 gave usable measurements), so it was really a prediction waiting for a large test.

The 350,757-flip test

František Bartoš and colleagues ran that test. 48 people flipped coins of 44 different currencies and denominations, caught each one in the hand, and recorded whether it landed on the side it started on. The flipping sessions were videotaped. The study appeared as a preprint in 2023 and was published in the Journal of the American Statistical Association in 2025 (journal version, free preprint). The main results:

• 178,079 of the 350,757 flips landed on the starting side: 50.8%, with a 95% credible interval of 50.6% to 50.9%. That is very close to the 51% prediction.
• Heads and tails overall were balanced at 50.0% heads. So if the starting side itself is random, the coin is fair.
• The size of the bias varied a lot from person to person, and some flippers showed little or none.
• The bias shrank as people flipped more. The authors suggest practice may have made their flips less wobbly.

The authors note one limitation: participants knew what was being tested, so deliberate influence can’t be fully ruled out, though they consider it unlikely.

What does 50.8% mean in practice?

Someone who could see the starting side and always called it would be right about 508 times in 1,000 instead of 500. In a single flip, that difference is impossible to notice.

It also explains why so many flips were needed. For n fair flips, the random wobble in a percentage (its standard error) is about 50 ÷ √n percentage points: about ±5 points for 100 flips, which hides a 0.8-point effect completely, but under 0.1 point for 350,757 flips.

Spinning a coin on a table is a different game

Diaconis and his co-authors note that coins spun on their edge can be far from fair, because the result depends on tiny details of the coin’s edge and centre of mass. They cite classroom data collected by Jim Pitman: 103 Berkeley students each tossed a penny 100 times and then spun the same penny 100 times. The tossed results looked like a fair coin. The spun pennies leaned clearly towards tails, and several students got fewer than 10% heads.

How the coin is handled	What the research says
Flipped and caught in the hand	Close to fair, with a slight lean (about 51%) towards the starting side
Caught, then slapped onto the back of the other hand	The extra turn reverses the lean, so it slightly favours the side opposite the start
Left to bounce on a table or the floor	Outside the model; the authors suspect a bouncing coin can spin briefly on its edge, which may make it less fair than a caught coin
Spun on its edge	Can be strongly biased, depending on the individual coin
Online coin	No physical starting side; 50/50 set by the code

Streaks are normal: the gambler’s fallacy

A fair coin has no memory. If it has landed heads four times in a row, the chance of heads on the next flip is still 1/2. The feeling that tails is “due” is known as the gambler’s fallacy.

The confusion comes from mixing up two questions. Before you start, the chance of five heads in a row is (1/2)⁵ = 1/32, about 3.1%. Once four heads have already happened, only the fifth flip is uncertain, and that is 1/2. Every specific sequence of five flips has the same 1/32 chance: HHHHH is exactly as likely as HTHTT (H for heads, T for tails). The run just looks more special.

Streaks of the same side, heads or tails, are much more common than most people expect:

Number of flips	Streak of the same side	Chance of at least one
10	4 or more	46.5%
20	5 or more	45.8%
50	6 or more	54.4%
100	7 or more	54.2%

These are exact figures: count every heads/tails sequence that has no streak that long, and subtract that share from 100%. A perfect split is also rarer than people think: in 100 flips, the chance of exactly 50 heads is only about 8%, although 40 to 60 heads happens about 96.5% of the time.

The current-streak counter on the Pick4Us coin is a record of what has happened, not a hint about what comes next.

How to use a coin flip well for a decision

1. Agree on the meanings first. Say out loud, or write down, which option is heads and which is tails, and whether the result is binding. Changing the rules after seeing the result is a quick way to make a flip feel unfair.
2. Call it in the air. With a real coin, one person flips and the other calls while the coin is in the air, without having seen the starting side. That way the small same-side lean helps nobody.
3. Catch it rather than spinning it. The research supports flipped-and-caught coins. Don’t settle anything by spinning a coin on a table.
4. Use best of three if one flip feels too abrupt. It doesn’t make the coin any fairer, since each side still wins exactly half the time, but a win feels more settled. A best of three takes 2.5 flips on average: half the time the first two flips match and it’s over, otherwise a third flip decides.
5. Notice your reaction. If you feel relief or disappointment as the coin lands, that tells you something about what you actually want. In two small lab studies, Jaffé, Reutner and Greifeneder (2019, PLOS ONE) found that when an on-screen coin (a die in the second study) pointed to one option, people’s feelings about the options became stronger, and many then decided by that feeling rather than simply following the coin. More in our article on decision psychology.

How the online coin works

The Pick4Us Flip Coin tool decides the result the moment you press Flip Coin. It asks the browser’s Math.random() for a number from 0 up to (but not including) 1. Below 0.5 means heads; anything else means tails. The 3D spin you see is an animation that lands on the result already chosen.

• There is no starting side, wobble or catch, so the same-side lean from the experiments doesn’t apply.
• The coin styles, Classic and Doge, change only the look. The same line of code decides every flip.
• The result panel shows how many flips you have made this session and your current streak of the same side.
• In Best of three mode, exactly two players take part: the left player scores on heads, the right player on tails, and the first to 2 points wins.
• The site cannot preset outcomes.

Math.random() is a pseudo-random number generator: perfectly good for deciding who goes first, but not designed for security. Our guide to how online random number generators work explains the difference.