Random Number Generator

What a random number generator returns

A random number generator takes a starting range and hands back a number from it, with every number in that range equally likely. Ask for a whole number between 1 and 100 and each of the 100 numbers has a 1 percent chance. Ask for a number between 1 and 6 and each face has a 1 in 6 chance, which is 16.7 percent. There is no memory in the tool, so a 7 that came up last time carries no weight into the next draw.

Three things make a draw meaningful, and this page handles all three:

• A stated range. The range is shown in the results, because a number without a range says nothing about its odds.
• A stated method. This page uses the browser's cryptographic source when it is available and says so in the output, and it can also replay a draw from a seed you type in.
• A stated count. How many numbers came out, and whether any were removed for being duplicates.

None of that makes a draw fair on its own. Fairness comes from the range and the rules being fixed before the draw, not from the quality of the number machine. A page that rolls a better random number still cannot fix a raffle where the organizer picks the winner by hand.

How the numbers are produced, and why Math.random is not enough

Browsers offer two sources of randomness, and they are not interchangeable. The Math.random() function that most web code reaches for is fast, decent for games and animations, and not built for anything where a person might want to predict the next result. The Web Crypto API exposes crypto.getRandomValues(), which is designed to survive an attacker who knows every previous output.

The MDN reference for getRandomValues notes one practical limit: a single call fills at most 65,536 bytes, which is 16,384 four byte numbers, so a large batch takes more than one call. The NIST SP 800-90A publication sets out the design rules for the deterministic generators that sit underneath these sources, and the W3C Web Crypto specification defines how browsers expose them to pages.

Seeded mode is the opposite of a security feature. A typed seed fixes the sequence, so anyone with the same seed and range gets the same numbers. That is what you want when the claim is we ran this draw once and here is the proof, and what you do not want for a password.

One more detail worth knowing: a raw random byte has 256 possible values, and dividing that by 6 does not divide evenly. 4 x 43 + 2 x 42 = 256, so a naive method makes four outcomes slightly more likely than the other two. Good generators avoid this by rejecting values that fall in the leftover part of the range and drawing again, which is what this page does.

Working out the odds before you draw

Most surprises with random draws come from arithmetic nobody did. Drawing 5 numbers from 1 to 100 sounds like 5 separate chances, but the chances shrink as you go, and the chance of a repeat grows fast.

The same idea drives the birthday problem. In a room of 23 people there is a 50.7 percent chance that two share a birthday, even with 365 days in the year, because 23 pairs is a lot of chances to land on the same date. The name comes from that surprise, not from a party. Drawing many numbers from a small range is the same calculation wearing different clothes.

For a raffle, the numbers that matter are per person rather than per draw. In a pool of 50 names, each person has a 1 / 50 = 0.02 chance of the first prize. After that name is removed the second prize has a 1 / 49 = 0.0204 chance and the third a 1 / 48 = 0.0208 chance, so each person ends up with a 3 in 50, or 6 percent, chance of taking one of the three prizes. A common mistake is to advertise the second prize as 1 in 50 too, which flatters the odds of the first winner and confuses the rest.

Worked example: a 50 person raffle with three prizes

Here is the whole calculation for a raffle with 50 entries and 3 prizes, drawn without replacement.

• First name drawn: each of the 50 entries has a 1 / 50 = 0.02 chance, which is 2 percent.
• Second name drawn: 49 entries remain, so each has a 1 / 49 = 0.0204 chance, which is 2.04 percent.
• Third name drawn: 48 entries remain, so each has a 1 / 48 = 0.0208 chance, which is 2.08 percent.

Every ordered sequence of three different names is equally likely, and there are 50 x 49 x 48 = 117,600 of them. Because the order of the three prizes does not matter for who wins what, the number of possible sets of three winners is 117,600 / 6 = 19,600, since each set can be ordered six ways. A single named person's chance of being in that set of 19,600 is 3 in 50, which is 6 percent, and their chance of being the first name is 2 percent.

Worked example two, a simpler one. Draw 5 numbers from 1 to 100 with no repeats allowed. The chance that all five are different is 100 x 99 x 98 x 97 x 96 / 100^5 = 0.9035, so 90.4 percent of draws come back with five distinct numbers and 9.7 percent contain a repeat. Each single number, 1 included, still has exactly a 1 percent chance of appearing somewhere in the list, the same as its chance on any one draw. That last fact is what makes the list fair even though the individual draws are not symmetric.

Now the case most raffles get wrong. If the raffle allows repeats, so the same person can win twice, then the first prize is still 1 / 50 = 0.02 per draw but a second prize can land on the same name, and the number of distinct winners drops below 3. The chance that all three prizes go to different people falls to 0.98 x 0.96 = 0.9408, so about 6 percent of draws produce fewer than 3 winners. Either rule is fair if it is published before the draw, and unfair only if the organizer switches to whichever rule gives a preferred result.

Picking from a list instead of a range

Sometimes the pool is already written down: names on a sign-up sheet, products in a warehouse, chores on a board. That case is a sample from a known set, and the odds work out the same way as the range case, with the set size replacing the upper bound.

Two choices matter, and both change the number you should quote:

Most raffles and most workload splits want no repeats, because a person who already won should leave the pool. The page reports which setting produced the result, so the rule is on the record.

When a draw has to be checked later, the audit trail matters more than the number. Write down four things: the range or the list, the count, whether repeats were allowed, and the seed if one was used. Those four lines are enough for someone else to rerun the same draw and get the same output, which is the point of a published seed. A screenshot of a result, on its own, proves nothing about how the result was produced.

For a bigger job, such as picking 30 of 900 names for a survey, the same generator handles it in one pass. Check that the list itself carries no order that matters: a sheet sorted by team or by signup time still gives every person an equal chance in a full shuffle, but drawing from the end of it does not.

When a seeded draw beats a one-off draw

Seeded mode answers a narrow question well: can you reproduce this result later? The seed is turned into a starting state for a small deterministic generator, and the sequence from that state is the same on every device and every visit. A demonstration of a shuffle, a reproducible example for a blog post, a test case that has to keep failing in the same way, or a draw that has to be replayed in front of an audience, all benefit.

Use a plain text seed such as a date, a ticket number, or a phrase. Keep it short enough to read aloud. The generator here mixes the characters of the seed before it starts, so a seed of 2026-10-23 does not behave like a nearby seed of 2026-10-24, which matters if seeds are being guessed at.

Two rules keep a seeded draw honest. Publish the seed before the draw, not after, or the seed is just a number that produces a result you liked. And keep seeds out of anything that has to stay secret: a seed that anyone can read is not a secret, and the password generator on this site is the right tool for secrets because it never stores or repeats a value you have seen.

Where a random number is the wrong tool

Some jobs need a random number and some need something else, and picking the wrong one causes real harm.

• Passwords and keys. Use the dedicated password generator or a password manager. A password built from Math.random or a visible seed is guessable, and reuse makes it worse.
• Medical doses and treatment decisions. Randomization belongs in clinical trials designed by statisticians with a written protocol, not in a web form.
• Legal and financial lotteries. Licensed operators use certified hardware, audited procedures, and published rules. A web page cannot offer that.
• Anything where a person can lose money. A draw you cannot explain to the loser is not a fair draw, whatever the number generator does.

For everyday use, a random number is a good tool for breaking a tie, choosing between equally good options, sampling a few items from a long list, splitting chores, picking a restaurant, or generating test data. For a quick dice-style roll with face counts and modifiers, the dice roller on this site is built for that. For a lottery with published odds and prize tiers, the lottery odds calculator covers the numbers.

What this page will not do is make a decision for you. It gives you a number from a range you set, under rules you set, and shows the odds of that range so the number means something. Whether the range was fair is still your call, and it is the call that matters most.