Put 23 people in a room and there is a 50.7 percent chance that at least two share a birthday, under the usual simplified assumptions.

That result feels wrong because 23 is tiny beside 365. But the comparison quietly asks the wrong question. We are not checking whether anyone matches one chosen birthday. We are checking every person against every other person.

With 23 people, there are 253 possible pairs. The surprising probability comes from all those opportunities to match.

Matching your birthday is a different problem

Suppose a person enters a room and asks whether anyone shares their birthday. Each other person has a 1 in 365 chance of matching that fixed date in the simplified model. Reaching a probability above 50 percent for that particular match would require 253 other people.

The birthday problem asks something much broader: does any pair in the room match?

Person one can match person two, three, four and everyone after them. Person two can match person three, four and everyone after them. Each new arrival creates several new pairs at once. That difference is the entire source of the apparent paradox.

Twenty-three people create 253 pairs

The number of distinct pairs in a group of n people is n multiplied by n minus one, then divided by two. For 23 people, that is:

23 × 22 ÷ 2 = 253 pairs

Each particular pair has a 1 in 365 chance of sharing a birthday under the model. Cornell University’s notes on probability in computer science use the same pair count to build intuition: the expected number of matching pairs is 253 divided by 365, or about 0.693.

That expected value is not the probability of at least one match. We cannot simply add 1/365 across all 253 pairs because the events overlap. If three people share one birthday, for example, that creates three matching pairs within a single room. The exact calculation needs a different route.

The clean calculation starts with no matches

It is awkward to count every possible way that at least two birthdays could coincide. There might be one pair, several pairs or three people on the same date. It is much easier to calculate the opposite event, in which every birthday is different, and subtract that probability from one.

The first person can have any birthday, so the chance that the group remains collision-free is initially 1. The second person must avoid one occupied date, giving a factor of 364/365. The third must avoid two dates, giving 363/365. The pattern continues until the twenty-third person must avoid 22 dates, giving 343/365.

P(no shared birthday) = (364/365) × (363/365) × ... × (343/365)

Multiplying those 22 fractions gives approximately 0.4927028. The complement is therefore:

P(at least one shared birthday) = 1 − 0.4927028 = 0.5072972

That is 50.72972 percent. Wolfram MathWorld gives the same exact formula and result.

Twenty-three is the first group to cross 50 percent

The threshold is close. With 22 people, the probability of at least one shared birthday is about 47.57 percent. Adding one person creates 22 new pairs, lifting the total above half.

The curve rises quickly after that. Under the same assumptions, the chance of a match is about 70.63 percent with 30 people, 97.04 percent with 50 and 99.92 percent with 70.

Cornell’s conditional-probability notes show the threshold directly: the chance that all birthdays differ is about 0.5243 for 22 people and 0.4927 for 23. Once the no-match probability falls below one-half, the match probability rises above it.

The standard model simplifies real birthdays

The textbook calculation assumes 365 possible birthdays, ignores February 29, treats every date as equally likely and assumes that one person’s birthday is independent of another’s.

Real birth dates are not distributed perfectly evenly across the calendar. Planned deliveries, holidays and seasonal patterns make some dates more common than others. Twins and other siblings also violate the simplest independence assumption if they are included together.

Uneven dates do not rescue the intuition behind the paradox. A perfectly uniform distribution spreads birthdays as widely as possible. Concentrating more births on some dates generally makes a collision at least slightly more likely. The 50.7 percent figure belongs to the clean 365-day model, while a calculation using real population data will depend on the country, years and people sampled.

The same mathematics appears in computing

The birthday problem is a small example of a broader collision problem. Computer systems often map a huge set of possible inputs into a smaller set of labels or hash values. Even if each output is individually unlikely, a collision becomes plausible after roughly the square root of the number of possible outputs has been sampled.

This is why computer-science texts teach the birthday calculation alongside hashing. A MIT Mathematics for Computer Science chapter moves directly from the 23-person result to hash collisions, while the US National Institute of Standards and Technology lists collision resistance as one of the defining requirements of approved cryptographic hash functions.

The security version operates on vastly larger spaces, but the logic is identical. A system can have far more possible labels than items and still encounter a duplicate earlier than linear intuition expects, because the number of pairs grows quadratically.

The paradox is really about human intuition

Nothing contradictory happens in the arithmetic. “Paradox” describes the gap between the answer and the estimate many people make before seeing the pairs.

Twenty-three people do not create 23 chances for a shared birthday. They create 253 relationships between pairs of birthdays. Once the room is viewed that way, 50.7 percent stops looking like a mathematical trick and starts looking like what it is: a compact lesson in how quickly opportunities for coincidence accumulate.