Dice pool probability for tabletop games: why the average roll lies to you

Ask a table of players what 3d6 averages and most will tell you "ten and a half" without hesitation. They're right. Ask them how often 3d6 actually rolls a 10 or 11 and the room goes quiet — because the average is the least interesting thing about a dice pool. The shape of the distribution is what decides whether your game feels swingy or steady, whether a +2 bonus matters, and whether your carefully chosen difficulty number is fair or impossible.

This is the part of game design that intuition gets wrong. Here's how dice pools actually behave, and how to reason about them before you put a number on a stat block.

The 30-second summary

• A single die is flat. Every face on a d20 is equally likely — 5% each. There is no "typical" roll.
• Adding dice creates a bell curve. 3d6 clusters hard around 10–11; the extremes get rare fast. More dice means a steeper, more predictable peak.
• A flat bonus shifts the whole curve. +2 moves every outcome up by two — it doesn't change the shape, only its position.
• Advantage isn't "+5ish". Its value depends entirely on the target number, and it's strongest in the middle of the range.

One die is flat. The number of dice changes everything.

Roll one d20 and every result from 1 to 20 has exactly a 1-in-20 chance. The distribution is a flat line. This is why d20 systems feel swingy: a level-12 fighter and a goblin both have a meaningful chance of rolling a natural 1 or a natural 20 on any given throw. The die doesn't know who's holding it.

Now roll 3d6. The lowest possible total is 3 (1-1-1) and the highest is 18 (6-6-6), but those are wildly unlikely — each happens once in 216 rolls, under half a percent. There are 27 different ways to roll a 10 and 27 ways to roll an 11, so those two totals together account for a quarter of all outcomes. The distribution isn't a line; it's a bell.

3d6 total	Ways to roll it	Probability
3 or 18	1 each	0.5% each
6 or 15	10 each	4.6% each
9 or 12	25 each	11.6% each
10 or 11	27 each	12.5% each

The lesson for designers: the more dice you add, the more the result behaves like "the average, usually." A system built on 3d6 (like GURPS) feels grounded and competent — heroes mostly perform at their skill level. A system built on 1d20 feels heroic and chaotic — anything can happen on any roll. Neither is wrong. They're different feelings, and you choose them by choosing how many dice go in the pool.

Why the average is the wrong question

"3d6 averages 10.5" is true and almost useless on its own. What you actually need to know when you set a difficulty of, say, 13 is: how often does the pool clear 13? For 3d6 the answer is about 26% — roughly one roll in four — even though 13 is only two and a half above the average. That's the bell curve at work: once you move past the peak, probability drops off a cliff.

This is exactly the kind of question a dice pool probability calculator answers directly. You build the pool, set the target, and read off the real probability of success instead of guessing from the mean. The gap between "two above average" and "one in four" is the gap that ruins encounters when designers eyeball it.

A flat bonus moves the curve without reshaping it

Add +2 to a 3d6 roll and you haven't changed the bell's shape at all — you've slid the entire thing two points to the right. The peak is now at 12–13 instead of 10–11. Every probability that applied to a total of N now applies to N+2.

This matters because it tells you what a bonus is worth. Against a target of 14, a 3d6 roller succeeds about 16% of the time. Give them +2 and they're effectively rolling against 12, which succeeds about 37% of the time. The same +2 against an easy target of 8 barely moves the needle — they were already succeeding 84% of the time. A flat modifier is most valuable when you're rolling near the middle of the curve and nearly worthless at the extremes.

What advantage actually does

"Roll twice, take the higher" — advantage in 5e and many other systems — is the most misunderstood mechanic at the table. People say it's "worth about +5." That's an average across all target numbers, and the average hides the real story.

Advantage gives you the most benefit when you have roughly a 50% chance to begin with. At that midpoint it's worth around +5 to your effective roll. But if you already succeed on a 2 or higher (95% chance), rolling twice barely helps — you were going to make it anyway. And if you need a natural 20, advantage roughly doubles your odds (from 5% to just under 10%), which sounds huge in relative terms but is still a small absolute gain.

• Easy tasks (need 5+ on d20): advantage adds only a couple of percent. You'd have made it regardless.
• Coin-flip tasks (need 11+): advantage is at its peak — roughly +5 to the effective roll.
• Near-impossible tasks (need 20): advantage nearly doubles a tiny chance. Big relative gain, small absolute one.

The practical takeaway for GMs: handing out advantage on a trivial check is a non-reward, and handing it out on a coin-flip check is genuinely powerful. Spend it where the odds are uncertain.

Exploding dice break the bell

Some systems (Savage Worlds, Shadowrun, many indie games) use exploding dice: roll the maximum face and you roll again, adding the new result, potentially forever. This puts a long, thin tail on the high end of the distribution. Most rolls behave normally, but every so often a die "aces" two or three times and produces a result far outside the nominal range.

The expected value of an exploding dN is its normal average multiplied by N/(N−1) — an exploding d6 averages 4.2 instead of 3.5, an exploding d4 averages about 3.3 instead of 2.5. The smaller the die, the more the explosions matter, because the maximum face comes up more often. The design effect is a system where heroes are usually competent but occasionally, unpredictably, spectacular. If that's the feel you want, exploding dice deliver it — but they make hand-calculation hopeless, which is exactly when you reach for a calculator that models the mechanic for you.

Setting fair target numbers

Once you understand the curve, difficulty design becomes deliberate instead of vibes-based. A rough framework for a pooled (bell-curve) system:

1. Decide the success rate you want a competent character to have. Routine tasks: 80–90%. Meaningful challenges: 50–65%. Long shots: 15–25%.
2. Find the target number that produces that rate for your pool plus the character's typical bonus. Don't reverse-engineer it from the average — read it off the actual distribution.
3. Check the extremes. Make sure your hardest tasks aren't literally impossible for an unskilled character, and your easiest aren't automatic for everyone.
4. Account for the swing. A flat d20 system needs a tighter band of target numbers than a 3d6 system, because the d20 already supplies the randomness.

Run each candidate pool and target through the dice probability calculator and you get the real success percentage, the mean, and the full histogram — including advantage, disadvantage, and exploding-dice variants. If you're pacing a whole session rather than a single roll, our encounter pacing tool works at the encounter level, and for card-based games the pack odds calculator handles the same probability questions for booster packs.

The one habit that fixes most dice-design mistakes

Stop reasoning from the average. The mean of a dice pool tells you where the curve is centred and nothing about how steep it is, and steepness is what players feel. A 13 against 3d6 and a 13 against 1d20+7 have the same average margin and completely different odds. Build the actual pool, look at the actual distribution, and set your numbers from there. The dice don't care what you intended — only what the curve allows.