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

By HSP · Published October 15, 2026 · 6 min read · Passwords & Randomness

People flip coins to settle who goes first, who picks the movie and which team kicks off. It feels like the fairest method there is. The short answer is that a coin flip is very close to fair, and the small bias that does exist is easy to remove once you know where it comes from.

Is heads or tails more likely?

Neither. In the large study described below, coins landed heads 50.0% of the time, with no sign that any particular coin favoured one face. So calling “heads” is not smarter than calling “tails”. What matters is not which face you pick, but which face was pointing up before the coin left the thumb.

The same-side bias, explained

In 2007, mathematicians Persi Diaconis, Susan Holmes and Richard Montgomery built a physics model of a human coin toss. A real flip is not a perfect end-over-end spin: the coin also wobbles slightly, a bit like a spinning plate. Because of that wobble, the face that started on top spends a little more time facing up during the flight. Their model predicted that a flipped coin lands on its starting side about 51% of the time.

In 2023 a team led by František Bartoš tested that prediction with 350,757 flips made by volunteers. They found:

• Same side as the start: 50.8% (the 95% range was 50.6% to 50.9%).
• Heads overall: 50.0%, so no heads-or-tails bias.
• Big differences between people: some flippers showed almost no bias, others more.
• The bias shrank as people flipped more, which fits the idea that practice gives a cleaner, less wobbly spin.

A 0.8 percentage point edge sounds tiny, and in a single toss it is. Over 1,000 tosses where you always know the starting face and call it, you would expect to win roughly 508 instead of 500. That will not change your life, but it does mean a hand toss is not mathematically perfect.

How to make a fair coin toss

1. Hide the starting side. Shake the coin in cupped hands, or let someone other than the caller place it on the thumb without showing it.
2. Call while the coin is in the air. The caller should not be able to see how it was loaded.
3. Flip high with plenty of spin. A low, lazy flip gives the coin fewer turns and makes the result easier to influence.
4. Agree on the rules first. Decide in advance whether you catch it or let it land, and what happens if it rolls away or is dropped.

Using a digital flip removes the physics completely. Our coin flip tool asks the browser for a random bit with the Web Crypto API, the same kind of source used to make strong passwords, so each result is an even 50/50 and nothing leaves your device.

Worked example: what 10, 100 and 1,000 flips look like

Even with a perfectly fair coin, results rarely split exactly in half. These probabilities come from the binomial formula, the standard way to count outcomes of repeated yes-or-no events:

So with 10 flips, an exact 5–5 split happens only about one time in four. A 6–4 or 7–3 split is completely normal. As the number of flips grows, the share of heads settles closer and closer to 50%, even though the raw gap between heads and tails can still grow. Try it: set the coin flip to 100 coins and run it a few times, then compare your totals with the table.

Streaks are normal, not a sign of a broken coin

Five heads in a row from a given starting point has a probability of 1 ÷ 32, about 3.1%. That sounds rare, but you get many starting points in a long sequence. In 20 fair flips, the chance of seeing a run of at least 5 of the same face somewhere is about 46%. In 100 flips, a run of 6 or more appears about 81% of the time.

This is why people who are asked to write down a “random-looking” list of heads and tails usually avoid long streaks, and why their lists are easy to spot. Real randomness is lumpier than our intuition expects.

The gambler’s fallacy

After five tails, many people feel heads is “due”. It is not. Each flip is independent: the coin has no memory, so the next toss is still about 50/50. The same logic applies to dice and lottery numbers. If you want to see this for yourself, roll the dice roller many times and notice that a long run of low numbers does not make a six more likely next.

A trick for a coin you do not trust

If you suspect a coin is bent or weighted, you can still get a fair result with a method credited to the mathematician John von Neumann:

1. Flip the coin twice.
2. Heads then tails counts as “A”. Tails then heads counts as “B”.
3. If both flips match (heads-heads or tails-tails), ignore them and flip twice again.

Why it works: if the coin shows heads with probability p, then heads-tails has probability p × (1 − p) and tails-heads has probability (1 − p) × p. Those are the same number, so A and B are equally likely no matter how biased the coin is. The cost is time, because matching pairs are thrown away.

When to use something other than a coin

A coin only gives two outcomes. For three or more choices, a random picker or a random number generator is fairer than a series of coin tosses, because a chain of 50/50 rounds can treat the choices unequally unless the number of options is a power of two. Our guides on generating random numbers and picking a random winner show how to run a fair draw step by step. If you enjoy probability puzzles, our page on brain games and skill tests has more to try.

Last updated: 15 Oct 2026. Educational overview only. Results above come from the cited study and standard probability formulas; individual flippers can differ.