Normal Approximation to Binomial Distribution
When the number of trials grows, the binomial formula turns into a long and tedious sum. The normal approximation to the binomial distribution replaces that sum with a single area under the bell curve — and with one small correction of ±0.5, the answer is usually right to two or three decimal places. This page explains when the shortcut is safe, how to use it in five steps, and checks every worked example against the exact binomial probability from our Binomial Probability Calculator.
The normal approximation to the binomial distribution says that when n is large and both np and n(1−p) are at least 5, a binomial B(n, p) behaves like a normal with mean μ = np and standard deviation σ = √(np(1−p)). Because the binomial is discrete and the normal is continuous, probabilities are adjusted by a ±0.5 continuity correction.
What Is the Normal Approximation to the Binomial?
A binomial distribution counts the number of successes, X, in n independent trials, each with the same success probability p. Its exact probabilities come from the binomial formula:
P(X = k) = C(n, k) × pk × (1 − p)n − k
That formula is exact, and it works for any n — but for large n it becomes impractical to use by hand. The probability of at least 60 heads in 100 coin flips, P(X ≥ 60), is a sum of 41 separate binomial terms. The normal approximation replaces the whole jagged histogram of the binomial with a smooth bell curve, and the area under that curve between two points estimates the probability. In symbols, if X follows a binomial distribution B(n, p), then for large n:
X ≈ N(μ, σ), where μ = np and σ = √(np(1 − p))
So the count X is approximately normal with mean μ = np and standard deviation σ = √(np(1 − p)). If the conditions in the next section hold, that single curve is all you need. The worked binomial distribution examples show what the exact formula looks like for small n, where it is still the right tool.
The idea is nearly 300 years old. Abraham de Moivre discovered the approximation in 1733, when he showed that the probabilities of a fair-coin binomial follow the bell curve as the number of flips grows. Pierre-Simon Laplace sharpened the result about fifty years later, extending it to any success probability p. The result is now known as the de Moivre–Laplace theorem, and it is the oldest special case of the central limit theorem — the reason a discrete count can be approximated by a continuous curve at all. The NIST/SEMATECH e-Handbook of Statistical Methods gives the same binomial distribution, with mean np and standard deviation √(np(1 − p)), as the formal definition.
When Can You Use the Normal Approximation to the Binomial?
Use the normal approximation when both np ≥ 5 and n(1 − p) ≥ 5. Both conditions must hold: np is the expected number of successes and n(1 − p) is the expected number of failures, and both need to be at least 5 for the binomial to be symmetric enough to look bell-shaped.
Some textbooks state the condition as np ≥ 5 and n(1 − p) ≥ 5; others are stricter and ask for np ≥ 10 and n(1 − p) ≥ 10. The difference is a matter of caution, not correctness: the larger the two counts, the closer the bell curve fits. Our Normal Approximation to Binomial Calculator checks the np ≥ 5 rule and shows a warning whenever either count falls below 5 — and Worked Example 5 below shows what happens when you ignore it.
Why do the conditions exist at all? A binomial with p close to 0.5 is nearly symmetric and starts to look like a bell even for modest n. A binomial with p near 0 or 1 is skewed — most of its probability piles up near one end — and the bell curve fits poorly until n is much larger. Even when the conditions pass, a skewed binomial converges slowly: with n = 100 and p = 0.05, np = 5 passes the lenient test, yet the approximation still misses by 0.0253, as the skewed-p case in “Why Is the Normal Approximation Sometimes Wrong?” shows. When in doubt, the stricter np ≥ 10 rule is the safer guide.
Before you approximate, run this check:
- Check np ≥ 5. The expected number of successes must be at least 5.
- Check n(1 − p) ≥ 5. The expected number of failures must be at least 5.
- If either check fails, do not approximate. Use the exact binomial probability instead.
What Is the Continuity Correction?
The binomial is discrete: it only takes whole-number values, and each value is a bar in a histogram. The normal is continuous: it is a smooth curve, and probabilities are areas under it. To use a continuous curve for a discrete count, each whole-number bar is treated as a unit-wide interval centred on the count — the bar for X = 6 runs from 5.5 to 6.5. Shifting every boundary by 0.5 is the continuity correction, and it shrinks the error sharply. In Worked Example 1 below, the correction moves the answer from 0.0228 to 0.0287, against the exact value 0.0284 — an error almost twenty times smaller.
The correction depends on the event. The four cases are:
| Binomial probability | Normal area (after correction) |
|---|---|
| P(X ≤ k) | Area up to k + 0.5 |
| P(X ≥ k) | Area from k − 0.5 |
| P(X = k) | Area from k − 0.5 to k + 0.5 |
| P(a ≤ X ≤ b) | Area from a − 0.5 to b + 0.5 |
Two habits make the correction easy. First, always draw the bar: “at most 8” includes the whole bar for 8, so the area goes up to 8.5; “at least 60” includes the whole bar for 60, so the area starts at 59.5. Second, remember that the correction matters most for small ranges of counts — for a single value such as P(X = 6), the uncorrected area is zero, and the correction is the whole answer.
How to Use the Normal Approximation in 5 Steps
Every normal approximation follows the same five steps:
- Check the conditions. Confirm np ≥ 5 and n(1 − p) ≥ 5. If either one fails, stop here and use the exact binomial probability.
- Find the mean and standard deviation. μ = np and σ = √(np(1 − p)).
- Apply the continuity correction. Move each boundary 0.5 towards the tail: k + 0.5 for “at most”, k − 0.5 for “at least”, and k ± 0.5 for a single value.
- Standardise. Turn the corrected boundary into a z-score: z = (x − μ) / σ. Our Z-Score Calculator does this step for any x, μ and σ.
- Read the area. Φ(z) is the area to the left of z under the standard normal curve; the area to the right is 1 − Φ(z). The Normal Distribution Calculator reads these areas directly, or the normal approximation calculator runs all five steps at once and shows the exact binomial beside the answer.
Worked Example 1 below follows the five steps in full. The examples after it use the same steps, so once you can do one, you can do them all.
Worked Example 1: P(X ≥ 60) in 100 Coin Flips
A fair coin is flipped 100 times. What is the probability of at least 60 heads? The exact answer is a sum of 41 separate binomial probabilities — P(X = 60) + P(X = 61) + … + P(X = 100) — exactly the kind of calculation the approximation exists for.
Step 1 — Check the conditions. np = 100 × 0.5 = 50 ≥ 5 and n(1 − p) = 100 × 0.5 = 50 ≥ 5. Both conditions hold, so the approximation is safe to use.
Step 2 — Mean and standard deviation. μ = np = 100 × 0.5 = 50. σ = √(np(1 − p)) = √(100 × 0.5 × 0.5) = √25 = 5.
Step 3 — Continuity correction. “At least 60” means 60, 61, …, 100 — and the bar for 60 runs from 59.5 to 60.5. So we need the area above 59.5, not the area above 60.
Step 4 — Standardise. z = (59.5 − 50) / 5 = 9.5 / 5 = 1.90.
Step 5 — Read the area. P(Z ≥ 1.90) = 1 − Φ(1.90) ≈ 1 − 0.9713 = 0.0287.
The approximation gives 0.0287. The exact binomial probability is 0.0284, so the absolute error is 0.0003 — the approximation is right to two decimal places. Without the continuity correction, z = 2.00 would have given 0.0228, an error almost twenty times larger. Check the working with the calculator tool and compare it with the exact binomial probability.
Worked Example 2: P(X = 6) with n = 20 and p = 0.3
A single value is the hardest case for a curve: a point on its own has no area under it. The continuity correction turns P(X = 6) into the area under the curve from 5.5 to 6.5.
Check the conditions. np = 20 × 0.3 = 6 ≥ 5 and n(1 − p) = 20 × 0.7 = 14 ≥ 5. The approximation is safe to use — note that np = 6 passes the lenient rule but not the stricter np ≥ 10 rule, so expect a slightly larger error than in Example 1.
Mean and standard deviation. μ = np = 20 × 0.3 = 6. σ = √(np(1 − p)) = √(20 × 0.3 × 0.7) = √4.2 ≈ 2.049.
Continuity correction. The bar for X = 6 runs from 5.5 to 6.5, so P(X = 6) becomes the area between those two boundaries.
Standardise. z1 = (5.5 − 6) / 2.049 ≈ −0.244 and z2 = (6.5 − 6) / 2.049 ≈ 0.244.
Read the area. Φ(0.244) − Φ(−0.244) ≈ 0.1927.
The approximation gives 0.1927. The exact binomial probability is 0.1916, so the absolute error is 0.0011 — about four times the error in Example 1, and still small. Check both values with the approximation calculator and the exact binomial value.
Worked Example 3: P(X ≤ 8) with n = 50 and p = 0.2
At most 8 successes in 50 trials, where each trial succeeds with probability 0.2 — for example, at most 8 faulty items in a batch of 50 when the fault rate is 20%.
Check the conditions. np = 50 × 0.2 = 10 ≥ 5 and n(1 − p) = 50 × 0.8 = 40 ≥ 5. Both conditions hold — and np = 10 even passes the stricter np ≥ 10 rule, so the approximation should be at its best.
Mean and standard deviation. μ = np = 50 × 0.2 = 10. σ = √(np(1 − p)) = √(50 × 0.2 × 0.8) = √8 ≈ 2.828.
Continuity correction. “At most 8” includes the whole bar for 8, so the area goes up to 8.5.
Standardise. z = (8.5 − 10) / 2.828 ≈ −0.530.
Read the area. Φ(−0.530) ≈ 0.2979.
The approximation gives 0.2979. The exact binomial probability is 0.3073, so the absolute error is 0.0094 — more than thirty times the error in Example 1, even though every condition is met and np = 10 passes the stricter rule. That is the honest limit of the method: the approximation is good, not perfect, and the gap is largest in the tails. Check the comparison in this calculator against the exact binomial answer.
Worked Example 4: Between 55 and 65 Heads in 100 Flips
For a range, the correction shifts BOTH ends: P(55 ≤ X ≤ 65) becomes the area from 54.5 to 65.5.
Check the conditions. np = 100 × 0.5 = 50 ≥ 5 and n(1 − p) = 50 ≥ 5. Both conditions hold.
Mean and standard deviation. μ = np = 100 × 0.5 = 50. σ = √(np(1 − p)) = √(100 × 0.5 × 0.5) = √25 = 5.
Continuity correction. The bar for 55 starts at 54.5 and the bar for 65 ends at 65.5, so the interval is 54.5 to 65.5.
Standardise. z1 = (54.5 − 50) / 5 = 0.90 and z2 = (65.5 − 50) / 5 = 3.10.
Read the area. Φ(3.10) − Φ(0.90) ≈ 0.9990 − 0.8159 = 0.1831.
The approximation gives 0.1831. The exact binomial probability is 0.1832, so the absolute error is 0.0001 — tied for the closest agreement of any example on this page. With two boundaries, the small errors at the two ends partly cancel. Check it in the normal approximation tool and the exact binomial calculator.
The Coin Flip Check: P(X ≤ 43) in 100 Fair Flips
Most textbooks work this one: at most 43 heads in 100 flips of a fair coin.
Check the conditions. np = 100 × 0.5 = 50 ≥ 5 and n(1 − p) = 50 ≥ 5.
Mean and standard deviation. μ = 50 and σ = 5, as in Examples 1 and 4.
Continuity correction. “At most 43” includes the whole bar for 43, so the area goes up to 43.5.
Standardise. z = (43.5 − 50) / 5 = −6.5 / 5 = −1.30.
Read the area. Φ(−1.30) ≈ 0.0968.
The approximation gives 0.0968. The exact binomial probability is 0.0967, so the absolute error is 0.0001. Verify the pair with the approximation tool and the binomial calculator.
Worked Example 5: When NOT to Approximate — n = 10, p = 0.1
The approximation has a gatekeeper: the conditions. With n = 10 and p = 0.1, np = 1 — far below 5 — so the calculator’s validity warning appears instead of a clean answer: “Approximation not recommended: np = 1.0 and n(1−p) = 9.0 — the normal approximation needs np ≥ 5 and n(1−p) ≥ 5.”
The parameters are μ = np = 10 × 0.1 = 1 and σ = √(np(1 − p)) = √(10 × 0.1 × 0.9) = √0.9 ≈ 0.949.
What happens if you ignore the warning and approximate P(X ≥ 2) anyway? The correction gives the area above 1.5, z = (1.5 − 1) / 0.949 ≈ 0.527, and 1 − Φ(0.527) ≈ 0.2991. The exact binomial probability is 0.2639, so the absolute error is 0.0352 — more than a hundred times the error in Example 1. The fix is always the exact binomial. Use the exact binomial instead: it gives 0.2639 with no approximation at all.
All five worked examples together:
| Example | n | p | Event | Approximation | Exact binomial | Absolute error |
|---|---|---|---|---|---|---|
| 1 | 100 | 0.5 | P(X ≥ 60) | 0.0287 | 0.0284 | 0.0003 |
| 2 | 20 | 0.3 | P(X = 6) | 0.1927 | 0.1916 | 0.0011 |
| 3 | 50 | 0.2 | P(X ≤ 8) | 0.2979 | 0.3073 | 0.0094 |
| 4 | 100 | 0.5 | P(55 ≤ X ≤ 65) | 0.1831 | 0.1832 | 0.0001 |
| Coin | 100 | 0.5 | P(X ≤ 43) | 0.0968 | 0.0967 | 0.0001 |
| 5 * | 10 | 0.1 | P(X ≥ 2) | 0.2991 | 0.2639 | 0.0352 |
* Example 5 fails the conditions, so its approximation should not be used at all.
Why Is the Normal Approximation Sometimes Wrong?
It goes wrong when the binomial is not bell-shaped enough: when np or n(1 − p) is small, when p is near 0 or 1, or when the probability covers only a small range of counts. In every case the fix is the same — use the exact binomial probability.
Skewed p. Take n = 100 and p = 0.05: np = 5 passes the lenient rule, yet the shape is badly skewed. For P(X ≤ 5), μ = 5 and σ = √(100 × 0.05 × 0.95) ≈ 2.18; the corrected z is (5.5 − 5) / 2.18 ≈ 0.23, giving 0.5907 against the exact 0.6160 — an error of 0.0253, more than eighty times the error in Example 1. A binomial with p = 0.05 approaches the normal much more slowly than one with p = 0.5, even when both pass the np ≥ 5 test. The stricter np ≥ 10 rule would have caught this one: np = 5 fails it.
Small ranges of counts. Even when the conditions are met, the approximation performs poorly for a small range of counts. A single count is one bar of the histogram, and the whole answer rests on the ±0.5 correction — Example 2 still came out well (error 0.0011), but narrow intervals of two or three counts can be much worse than the broad ranges in Examples 1 and 4.
Small n. When n itself is small, the histogram has few bars and no bell curve can fit it. That is the warning case in Example 5: np = 1, and the approximation misses by 0.0352.
The Central Limit Theorem Connection
The normal approximation is the central limit theorem at work. Each trial is a Bernoulli outcome — 1 for a success, 0 for a failure — and the binomial count X is the sum of n independent Bernoulli variables. The central limit theorem says that sums of independent variables tend to a normal distribution as n grows, whatever the shape of the individual variables. That is why the approximation improves as n grows, and why it struggles when p is near 0 or 1: a very skewed Bernoulli variable needs many more trials before its sum looks normal. The de Moivre–Laplace theorem is the special case of the central limit theorem for Bernoulli trials, proved about 180 years before the general theorem.
Common Mistakes to Avoid
- Skipping the conditions check. The approximation is only as good as its conditions. Check np ≥ 5 and n(1 − p) ≥ 5 before you start — and remember the stricter np ≥ 10 rule when p is far from 0.5.
- Forgetting the ±0.5 correction. Without it, Example 1 gives 0.0228 instead of 0.0287 — an error almost twenty times larger.
- Correcting the wrong way round. For “at most” the boundary moves up (k + 0.5); for “at least” it moves down (k − 0.5). Getting them backwards makes the error worse than no correction at all.
- Treating the approximation as exact. It is an estimate with a real error — 0.0094 in Example 3. Always compare it with the exact binomial when the answer matters.
- Using it when np < 5. The calculator warns you for a reason. With np = 1, as in Example 5, the approximation is not just imprecise — it is the wrong tool.
Frequently Asked Questions About the Normal Approximation to the Binomial
When can you use the normal approximation to the binomial?
When both np ≥ 5 and n(1 − p) ≥ 5. The two counts are the expected numbers of successes and failures; both need to be at least 5 for the binomial to be symmetric enough to look bell-shaped. Some textbooks use the stricter rule np ≥ 10 and n(1 − p) ≥ 10, and our calculator checks the np ≥ 5 rule and warns you whenever either count falls below 5.
What is the continuity correction?
The continuity correction is the ±0.5 shift used when a discrete binomial probability is estimated from the continuous normal curve. Each whole-number bar of the binomial histogram is treated as a unit-wide interval — the bar for X = 6 runs from 5.5 to 6.5 — so P(X ≤ k) becomes the area up to k + 0.5, P(X ≥ k) the area from k − 0.5, and P(X = k) the area from k − 0.5 to k + 0.5.
Why is the normal approximation sometimes wrong?
It goes wrong when the binomial is not bell-shaped enough: when np or n(1 − p) is small, when p is near 0 or 1, or when the probability covers only a small range of counts. The approximation is good, not perfect — Example 3 on this page misses by 0.0094 even though every condition is met — and the fix is always the exact binomial probability.
How do you find the mean and standard deviation for a binomial?
The mean is μ = np and the standard deviation is σ = √(np(1 − p)). For example, with n = 100 and p = 0.5, μ = 100 × 0.5 = 50 and σ = √(100 × 0.5 × 0.5) = 5. These are the two numbers every normal approximation starts from.
Do you add or subtract 0.5 for P(X ≥ k)?
Subtract. P(X ≥ k) includes the whole bar for k, and that bar starts at k − 0.5, so the area runs from k − 0.5 upwards. For P(X ≤ k) you add: the bar for k ends at k + 0.5, so the area runs up to k + 0.5. Getting them backwards makes the error worse than no correction at all.
What is the z-score for the normal approximation?
After the continuity correction, the z-score is z = (x − μ) / σ, where x is the corrected boundary, μ = np and σ = √(np(1 − p)). For example, for P(X ≥ 60) with n = 100 and p = 0.5, the corrected boundary is 59.5, μ = 50 and σ = 5, so z = (59.5 − 50) / 5 = 1.90, and P(Z ≥ 1.90) ≈ 0.0287.
When should you use the exact binomial instead?
Whenever np or n(1 − p) is below 5, whenever p is near 0 or 1 and the shape is badly skewed, whenever the answer must be exact rather than approximate, and whenever n is small enough for the exact formula to be painless. The exact binomial also has no continuity correction to get wrong. Our Binomial Probability Calculator gives the exact probability for any n, p and event.
Is the normal approximation the same as the central limit theorem?
No — the approximation is the central limit theorem applied to one special case. Each trial is a Bernoulli outcome, 1 or 0, and the binomial count is the sum of n independent Bernoulli variables. The central limit theorem says such sums tend to a normal distribution as n grows, and the de Moivre–Laplace theorem is that statement for Bernoulli trials, proved about 180 years before the general theorem.
Conclusion
The normal approximation to the binomial distribution turns an impossible-looking sum into a single area under a bell curve. Check that np and n(1 − p) are both at least 5, take μ = np and σ = √(np(1 − p)), shift each boundary by 0.5, standardise, and read the area. When the conditions hold, the answer is usually good to two or three decimal places; when they do not, the exact binomial is one click away. Try it yourself: run Example 1 through the normal approximation to binomial distribution calculator — it shows the approximation 0.0287 next to the exact 0.0284, and the 0.0003 gap between them is the whole story in one number.
Check Tools
Put the examples above into practice with these free tools.
Normal Approximation to Binomial Calculator
Use the normal approximation with a continuity correction when n is large and the binomial formula gets heavy.
ToolsTry Calculator → ● LiveBinomial Probability Calculator
Work out exact and cumulative binomial probabilities and check every example on this page in seconds.
ToolsTry Calculator → ● LiveNormal Distribution Calculator
Find exact probabilities, percentiles, and z-scores for any normal distribution — the other half of the approximation.
ToolsTry Calculator → ● LiveExplore All Statistical Tools
Browse the full set of free statistics calculators and generators in one place.
ToolsBrowse All Tools →