Is a Coin Flip Really 50/50? What the Science Says

What physics predicted: a slight bias toward the starting side

In 2007 the mathematicians Persi Diaconis, Susan Holmes and Richard Montgomery published a physical analysis of the coin toss (Dynamical Bias in the Coin Toss, SIAM Review). Their model concluded that a coin flipped in the usual way, spinning on its axis with some wobble, tends to land on the same side that was facing up at the start, with an estimated probability of around 51%.

The intuition is that when the coin spins and wobbles at once, it spends a slightly larger fraction of its time in the air with the starting face upward. The effect is small because human flips spin very fast and vary a lot, which is why the model predicted a bias close to 1%, not 10 or 20%.

What was measured: 350,757 flips

That prediction was tested in 2023 in a large-scale experiment: a team led by František Bartoš collected 350,757 flips of real coins and recorded how each one landed. The coin landed on the same side it started on 50.8% of the time, with a 95% credible interval between 50.6% and 50.9%, very close to what the model predicted. They found no advantage between heads and tails: each came up in 50.0% of flips.

They also found that the bias varies a lot from person to person: people who flip with more wobble show more bias toward the starting side, and in some people it approached 60%. The study concludes the effect is real but small.

What it means in practice

A 50.8% bias means that out of 1,000 flips you would expect about 508 to land on the starting side instead of 500. To decide who pays for coffee, that is irrelevant. To settle a tie between two people, what best protects fairness is a simple procedure: one person flips the coin and the other calls heads or tails while it is in the air. That way the caller cannot take advantage of knowing which side started facing up.

On the edge: it has been estimated that a coin can land on its edge roughly once in 6,000 flips in one specific case (a US nickel). That is an estimate from a physical model, not a measured frequency, and it depends heavily on the coin and the surface. In a normal flip you can ignore it.

Streaks don't mean the coin is "rigged"

Many people distrust a coin when five heads come up in a row, but long streaks are normal in a random sequence. In 20 flips, the probability of seeing a run of at least 5 identical results is about 46%. In 100 flips, it is about 81% for a run of at least 6, and about 54% for a run of at least 7. In fact, a sequence with no long runs looks suspicious because it usually means someone tried to "make it look random".

And every flip is independent: after five heads in a row, the probability of heads on the next one is still 50%. Believing it is "due" for tails is the gambler's fallacy. The coin has no memory.

The virtual coin: no wobble, no starting side

A virtual coin has no physics. The starting-side bias depends on the motion of a real object, so it cannot exist in a generator. What does matter is the source of randomness: a common pseudo-random generator (like Math.random() in JavaScript) can have an internal state that can be reconstructed from a few outputs, whereas a cryptographic generator backed by the operating system cannot be predicted.

This site's coin flip uses crypto.getRandomValues(): each flip gets a cryptographically secure random value and converts it into heads or tails in a way that gives both exactly the same probability, with no modulo bias. Each flip is independent of the previous ones and does not depend on who makes it.

When to use a coin (physical or virtual)

It is useful for settling ties or binary decisions where neither option is clearly better and what matters is that everyone accepts the result: who starts a game, who chooses first, informal tiebreakers. For decisions with more than two options, the wheel or the random number generator are better. And for decisions involving money or serious consequences, set the rules in advance and keep a record of the result.

Frequently asked questions

Is a physical coin perfectly fair?

Almost. An experiment with 350,757 flips found it lands on the same side it started on 50.8% of the time. Between heads and tails it found no difference. The bias is real but very small.

Can I win by always betting on whichever side is up before the flip?

In theory you would have an edge of less than one percentage point, which varies by who flips. In practice it is not a useful strategy.

Does the virtual coin have the same bias?

No. That bias comes from the physics of a real coin spinning, and a generator has no physical motion. With a cryptographically secure source, each flip has exactly a 50% chance of heads and 50% of tails.

Five heads in a row came up. Is the coin rigged?

Not necessarily. Long streaks are normal: in 100 flips a run of at least 6 identical results is more likely than not. Each flip is still independent.

Can a coin land on its edge?

Yes, but it is extremely rare. A physical model has estimated around 1 in 6,000 for one specific coin; it is an estimate, not a measurement.