Mathematics & Logic · Probability & statistics
The normal distribution and the bell curve
Measurement errors, dice totals and test scores often pile up in the same bell shape. See why many small effects added together produce it, how to read it with the 68-95-99.7 rule, and when assuming it goes wrong.

What you will learn
- Explain why many small random pushes, added together, pile up in the middle
- Say which kinds of measurement tend to be bell-shaped and the condition behind it
- Explain how the mean and standard deviation set a bell curve's center and width
- Use the 68-95-99.7 rule to estimate what share of values falls in a range
Lesson 1 of 4 · 4 min
Why random bounces pile up into a bell
A Galton board is an upright board with rows of pegs and a row of slots at the bottom. Drop a bead in at the top. At each peg it bounces left or right, with even chances. After the last row it lands in a slot.
Pour in thousands of beads and they don't spread evenly. They stack high in the middle slots and thin out toward the edges, in a bell shape. Why?
Each bead's slot depends on how many times it went right. To reach the far left slot, it has to go left at every single peg. There's only one path for that. To reach the middle, it needs about as many lefts as rights, in any order, and there are many such paths.
The board is a small model of a big pattern. When a result is the sum of many small effects that act independently, its values pile up in a bell. Each effect is like one bounce. Most of them cancel, and a few runs pile up in one direction.
The total of many dice works this way. So do errors when you measure something carefully. A reading is nudged up or down by many small things, such as hand shake, temperature and how you read the scale. In 1809 the mathematician Carl Friedrich Gauss used the bell as a model for measurement errors.
The bell shape is called the normal distribution. It only appears when the effects add up and none of them dominates.
Try to recallWhy are careful measurement errors often bell-shaped, and what condition does that need?
Each reading is pushed up or down by many small, independent effects, and they add up. Most pushes cancel, so readings cluster near the true value and big errors are rare. That only works when the effects add and no single one dominates.
A bell curve appears when many small, independent effects add together, because there are far more ways for them to cancel than to agree. You can now explain the Galton board and say why measurement errors follow the same shape.
Lessons in this topic
Why random bounces pile up into a bell
Drop beads through a board of pegs and they stack into a bell. See why adding many small random pushes does that, and which real measurements come out bell-shaped.
Reading a bell curve with the 68-95-99.7 rule
Two numbers describe any bell curve, its center and its width. Learn to use them to say what share of values falls in any range.
Why averages come out bell-shaped
Even lopsided or flat data give bell-shaped averages. See what the central limit theorem says, why averages of bigger samples cluster more tightly, and its limits.
When the bell curve gets the world wrong
Incomes, stock prices and market crashes break the bell. See what happens when effects multiply, and why fat tails make extreme events far more common than a bell allows.
Three quick stories from this topic
Regretting the name normal
The statistician who made 'normal curve' the usual name later regretted it
In 1920 Karl Pearson wrote that the name led people to believe every other distribution was abnormal. Plenty of real data, such as incomes, are lopsided rather than bell-shaped.
Why 'normal'
By the late 1800s some writers called the curve normal because it seemed typical and common.
Another name
It's also called the Gaussian distribution, after Carl Friedrich Gauss, who used it to model errors in measurement.
Beads find the middle
On a 10-row pegboard, a quarter of falling beads land in the middle slot of 11
Each bead bounces left or right at random at every peg. Of the 1,024 possible paths, 252 end in the middle and just one ends at each edge.
Why the middle wins
Lefts and rights cancel out far more often than they all agree, so most beads end near the center.
A bell from bounces
With many beads, the stacks form a bell curve. The same thing happens whenever many small, independent effects add up.
The impossible crash
On a bell curve, a six-standard-deviation day should come once in 1.4 million years
Yet on 19 October 1987 the Dow Jones index fell 22.6% in a single day. Real market moves have fat tails, so extreme days come far more often than a bell allows.
Thin tails
A bell's chances fall off very fast away from the mean. That's what makes huge moves look nearly impossible on paper.
An old warning
In the early 1960s, Benoit Mandelbrot found that daily cotton price changes going back to 1900 didn't follow a bell.
Sources
- Galton board, Wikipedia
- Normal distribution, Wikipedia
- Central limit theorem, Wikipedia
- 68-95-99.7 rule, Wikipedia
- Black Monday (1987), Wikipedia
- Fat-tailed distribution, Wikipedia





