What Is the Average Value of a Dice Roll?

The average, or expected, value of one roll of a fair six-sided die is 3.5, computed as (1+2+3+4+5+6) ÷ 6. For n dice summed together, the expected total is n × 3.5, so two dice average 7.0 and five dice average 17.5 — numbers that describe a long-run average, not a value either roll can actually show.

A worked example with numbers

For any fair die numbered 1 through s, the expected value is (s + 1) ÷ 2:

How to set the rule before the result

To find the expected value of a die with non-standard face values — the kind produced by the custom dice roller — sum every listed face value and divide by the number of faces, rather than assuming consecutive integers starting at 1. A die with faces 2, 2, 4, 4, 6, 8 has expected value (2+2+4+4+6+8) ÷ 6 = 26 ÷ 6 ≈ 4.33, even though its highest face is 8. For a dice pool that sums several dice, add each die’s own expected value; for a pool that counts successes above a threshold instead of summing, expected value uses the success probability, not the face numbers.

Common mistakes that change the odds or the process

A frequent mistake is expecting a short session to land near the average: five rolls of a d6 can easily total 22 (average 4.4) or 12 (average 2.4) without anything being wrong, because the long-run average only becomes visible over many more trials. A second mistake is assuming expected value equals the most likely single outcome — true for a symmetric distribution like 2d6, but false for a skewed die such as one with faces 1, 1, 1, 1, 1, and 20, whose expected value is (5 + 20) ÷ 6 ≈ 4.17 even though 1 is by far the single most common result. A third is confusing the expected value of one die with the expected value of a dice pool’s full sum. A fourth, subtler mistake shows up when a game reports an “average roll” from a mix of different die types without weighting by how often each type is used — averaging 3.5 (a d6) and 10.5 (a d20) as if they were two equally weighted samples gives 7.0, which does not describe either die and is not a meaningful statistic unless the two dice are actually used equally often in practice.

Where this method stops being appropriate

Expected value assumes a fair, undamaged, correctly numbered die and independent rolls; it says nothing about the outcome of any one roll or a short streak, and it does not apply unchanged to a die that is weighted, chipped, or read by an unreliable method. It also does not extend automatically to a dice-and-card system where later draws depend on earlier ones, such as a deck that is not reshuffled between rounds — that violates the independence the average calculation assumes.

How the random source fits into the rule

Wolfram MathWorld’s dice reference, Wolfram MathWorld’s dice reference, documents the arithmetic-mean definition used for a fair die’s expected value and the combinatorics behind multi-die sums, including why 2d6 concentrates around 7 while a single d12 spreads its probability evenly across 1 through 12.

How to find the expected value of a weighted or custom die

The formula is always the same: add every face’s value once, then divide by the number of faces. It does not matter whether the faces repeat, skip numbers, or use large gaps, as long as each face is equally likely to land — a die that is not physically balanced breaks that assumption regardless of what the calculation says.

The same formula extends to a dice pool that sums several different custom dice together: add each die’s own expected value to get the pool’s expected total. A pool of one d4, one d6, and one d8 has expected value 2.5 + 3.5 + 4.5 = 10.5, whether or not the three dice share the same number of faces. This additivity holds regardless of how skewed any individual die’s face values are, which is why it is a faster route to a pool’s average than trying to enumerate every combination of results by hand.

Why the average value cannot predict your next roll

Expected value is a property of the whole probability distribution, not a forecast for an individual event. A d6 with expected value 3.5 is exactly as likely to show 1 as to show 6 on the very next roll; the 3.5 figure only becomes a useful description of a large number of rolls added together and divided by the count. Treating it as a target — “a higher number is due” — is the same reasoning error covered in why repeated results do not prove a random picker is broken, applied to dice instead of a list picker.

A useful check is to record ten real d6 rolls and total them: a sum anywhere from roughly 25 to 45 is unremarkable, even though the “expected” total for ten fair rolls is exactly 35. Only once a session runs into the hundreds of rolls does the running average reliably settle close to 3.5 per die, which is the sense in which expected value is a long-run description rather than a short-run prediction.

The same idea explains why a game that rewards “above-average” dice results over a short session, such as a single evening of play, cannot reliably reward genuine skill through the dice alone: with a small number of rolls, deviations of a full point or more from the theoretical average are common for entirely fair dice, so a short-session bonus tied to a raw average mostly measures luck rather than any property of the player or the equipment.

How spread out each die really is, not just its average

Two dice can share the same expected value while spreading their results very differently. A d6 and a lopsided custom die with faces 1, 1, 1, 1, 1, 20 can be built to land near the same mean, yet the d6’s standard deviation of about 1.71 is far smaller, because the custom die’s single 20 pulls results away from the middle much harder than any d6 face can. Expected value alone never says how far a typical roll strays from that average — the standard deviation column does.

A second case: the expected value of 4d6 drop lowest

Rolling four d6 and discarding the lowest before summing the rest — the common method for generating a tabletop ability score — has an exact expected value of about 12.24, not the 10.5 a naive shortcut might suggest by taking 4 × 3.5 = 14 and subtracting one die’s average of 3.5. That shortcut is wrong because the discarded die is not a random, unbiased one: it is specifically whichever die rolled lowest, so it strips away more probability mass from the low end than an average die would. Enumerating all 1,296 possible four-die outcomes gives an exact expected value of 12.2446 and a standard deviation of about 2.85, both measurably higher than plain subtraction predicts.

The “I’m due for a high roll” mistake, worked with numbers

After five low rolls in a row on a fair d6, the sixth roll is not more likely to be high. Suppose a player rolls 1, 2, 1, 3, 2 across five turns — a below-average run — and reasons a 6 is “due.” Each roll stays an independent trial with P(6) = 1/6 ≈ 16.67%, exactly what it was before the low run started. The correct way to describe the streak is retrospective, not predictive: five rolls averaging 1.8 against a long-run average of 3.5 is an unremarkable dip for five trials, one that fades only because many future independent rolls dilute it, not because the die adjusts itself to compensate.

How many rolls it takes to catch a 1-percentage-point biased face

Catching a die that favours one face by as little as one percentage point above its fair 16.67% share — say a 6 landing 17.67% of the time instead of 16.67% — with a 95% confidence interval no wider than ±1 percentage point, needs roughly n = 1.96² × (1/6)(5/6) ÷ 0.01² ≈ 5,336 logged rolls of that one die. Fewer rolls cannot reliably separate a genuinely biased face from ordinary short-run variation; a home test of 50 or 100 rolls, however suspicious the run looks, is far too small a sample to draw a conclusion from.

Where the average stops describing the roll

Expected value describes a long-run average under independent, identically distributed rolls of a fair die; it says nothing about a single roll, a short session, or a die that is chipped, filed, or otherwise physically unbalanced. It also breaks down the moment rolls stop being independent — a card-based “dice replacement” drawn from a deck that is not reshuffled between rounds changes its own odds as cards are removed, so the plain arithmetic-mean formula for a fair die does not apply to it without modification.

Related DiceDecide tools

• Roll two dice and see the sum distribution
• Build a die with custom face values
• Read why two-dice sums are not equally likely