How Many Different Ways Can You Split a Group Into Two Even Teams?
A group of 12 people split into two labeled teams of 6 has exactly C(12,6) = 924 distinct possible Team A rosters, each pairing with one specific, fully determined Team B roster of the remaining 6 people. If the two teams are not separately labeled — just “two groups” rather than a named Team A and Team B — that count halves to 462, because swapping which group is called “A” and which is called “B” produces the same underlying split counted twice in the labeled figure. The same binomial-coefficient formula, C(n, n/2), scales to any even group size: 4 people give 6 labeled splits (3 unlabeled), 8 people give 70 (35 unlabeled), 20 people give 184,756 (92,378 unlabeled), and 30 people give over 155 million labeled splits.
A worked example with numbers
For a group with an odd number of people, an exactly even split is impossible, but the same formula still counts the nearest even split: 11 people dividing into groups of 6 and 5 has exactly C(11,5) = C(11,6) = 462 distinct ways to choose which 5 (or 6) people form the smaller group, with the remaining people automatically forming the other. Splitting into more than two groups uses a direct extension of the same idea: dividing 12 people into three labeled teams of 4 each has 12! ÷ (4!×4!×4!) = 34,650 distinct labeled arrangements, or 34,650 ÷ 3! = 5,775 if the three teams are otherwise unlabeled and interchangeable, since any of the 6 ways to permute three equal-sized team labels among themselves produces the same underlying three-way split.
How to set the rule before the result
To generate one of these splits fairly and uniformly at random — not just to know how many exist — shuffle the full roster with a genuinely unbiased shuffle, such as the one behind the list shuffler, then take the first half of the shuffled order as Team A and the rest as Team B. This correctly gives every one of the C(n, n/2) labeled splits an exactly equal 1 ÷ C(n, n/2) chance, provided the underlying shuffle itself is unbiased; a shuffle with a subtle implementation flaw can silently favor some splits over others without changing how many splits are theoretically possible. It is worth explicitly checking that a “random teams” tool is actually doing this shuffle-then-slice approach, rather than a flawed shortcut such as flipping an independent coin for each person, which does not reliably produce an even split at all.
Common mistakes that change the odds or the process
That flawed shortcut is worth quantifying directly: flipping one independent fair coin per person and sorting Heads onto Team A only produces an exactly even 6-and-6 split for a 12-person group about 22.56% of the time — C(12,6) ÷ 2¹² = 924 ÷ 4,096 — meaning roughly 3 out of 4 attempts using this method would produce an uneven team, such as 7-and-5 or worse, not the even split most people assume a per-person coin flip guarantees. The gap widens as the group grows: a 20-person group split by independent per-person coin flips lands on an exact 10-and-10 split only about 17.6% of the time. A second common mistake is confusing the labeled and unlabeled counts when reading a claim like “X possible team combinations” — most tools and most casual usage mean the labeled figure, since a captain-assigned Team 1 and Team 2 are treated as distinct rosters, so quoting the unlabeled figure in that context understates the real count by exactly half. A third mistake is assuming the three-or-more-team formula is a simple extension of squaring or tripling the two-team formula, when it actually requires the full multinomial coefficient divided by the number of ways the equal-sized team labels can be permuted among themselves, a genuinely different calculation from the two-team binomial case.
Where this method stops being appropriate
These counts describe distinct possible splits, not the probability that any particular split is competitively balanced by skill — a uniformly random split treats every roster as equally likely regardless of ability, the separate distinction this site’s guide to random versus balanced teams covers directly. This formula also assumes the two (or more) team sizes are meant to be equal; splitting a group into deliberately unequal sizes, such as 7 and 5, uses a single labeled C(n, k) count with no automatic halving for an unlabeled version, since swapping which physical group is called “first” also swaps which one has 7 members and which has 5, a different, non-symmetric situation from an equal split.
How the random source fits into the rule
NIST’s binomial-distribution reference documents the binomial coefficient this counting is built on, the same C(n, k) formula used throughout this site wherever a fixed number of items must be chosen from a larger set.
Ways to split a group into two even teams, by group size
| Group size | Labeled splits (Team A vs Team B) | Unlabeled splits (two interchangeable groups) |
|---|---|---|
| 4 | 6 | 3 |
| 6 | 20 | 10 |
| 8 | 70 | 35 |
| 10 | 252 | 126 |
| 12 | 924 | 462 |
| 20 | 184,756 | 92,378 |
| 30 | 155,117,520 | 77,558,760 |
The unlabeled column is always exactly half the labeled column, for every group size shown, because swapping the two team labels is the only symmetry an even two-way split has.
A second case: splitting into three even teams instead of two
Dividing the same 12-person group into three labeled teams of 4 gives 12! ÷ (4!×4!×4!) = 34,650 distinct arrangements — far more than the 924 two-team splits, because there are more ways to distribute people across three groups than across two. If the three teams have no names, only 34,650 ÷ 3! = 5,775 of those arrangements are genuinely distinct, since every unlabeled three-way split can be labeled Team A/B/C in 3! = 6 different orders that all describe the same underlying grouping. The same division-by-k! rule applies to any number of equal-sized, unlabeled teams: divide the raw multinomial count by the factorial of how many interchangeable teams there are.
Why flipping a coin per person does not reliably split a group evenly
Assigning each of 12 people to a team with an independent coin flip treats every one of the 2¹² = 4,096 possible Heads/Tails outcomes as equally likely, but only 924 of those 4,096 outcomes happen to produce an exact 6-and-6 split — the rest land somewhere else, from a mild 7-and-5 imbalance to, in principle, an extreme 12-and-0 result. That gives a 924 ÷ 4,096 ≈ 22.56% chance of an exactly even split for 12 people, dropping to roughly 17.6% for 20 people, which is why any tool that promises “random but even” teams needs a shuffle-and-slice method rather than independent per-person coin flips — the two approaches are both genuinely random, but only one of them guarantees the even part.
How likely two specific people are to end up on the same team
Out of the 924 labeled ways to split 12 people into two teams of 6, exactly how many keep two named people, say Alice and Bob, on the same side? Counting only the splits where both land on Team A: the other 4 Team A spots are chosen from the remaining 10 people, giving C(10,4) = 210 such splits; the same count, 210, applies to splits where both instead land on Team B. Adding the two together gives 420 of the 924 labeled splits — about 45.45% — keeping Alice and Bob together, only slightly under the naive 50% guess a person might reach without working through the combinatorics, and a useful sanity check for anyone building or auditing a “random teams” tool that also needs to report how often two specific people should expect to be paired. The same 45.45% figure also answers a related, commonly asked question about siblings or friends placed in the same randomly split group: it is a genuinely common, not a rare, outcome, which is worth knowing before treating a single instance of it as evidence that a split was not truly random.