Binomial vs normal distribution

By Jude Wallis · Published

A binomial distribution is discrete: it gives the probability of each whole-number count of successes in nn independent trials with success probability pp. A normal distribution is continuous: a symmetric bell curve set by a mean μ\mu and standard deviation σ\sigma, with probabilities as areas.

AP Statistics: Unit 2 (topics 2.10 The Binomial Distribution, 2.11 The Normal Distribution). In the Fall 2026 AP Statistics course, the binomial distribution is topic 2.10 and the normal distribution is topic 2.11, back to back in Unit 2, which is 15% to 25% of the multiple-choice section. The large counts check that links them returns in Unit 3 as a condition for inference about proportions.

Binomial vs normal distribution: the short answer

The two distributions model different kinds of random variables.

  • A binomial random variable is a count. It answers "how many successes in nn trials", so its values are the whole numbers 0 through nn and nothing in between.
  • A normal random variable is a measurement on a continuous scale, such as a height or a time. It can take any value in an interval.

That split drives everything else. A binomial probability is the height of a bar at one count, found from a formula. A normal probability is an area under a curve between two values. A binomial distribution is set by nn and pp. A normal distribution is set by μ\mu ("mu", the mean) and σ\sigma ("sigma", the standard deviation).

What a binomial distribution is

A random variable XX is binomial when it counts successes under four conditions: a fixed number of trials nn, two outcomes per trial (success or failure), the same success probability pp on every trial, and independent trials. The check is covered in how to tell if a situation is binomial.

The probability of exactly xx successes is

P(X=x)=(nx)px(1−p)n−xP(X = x) = \binom{n}{x} p^x (1-p)^{n-x}

with mean and standard deviation

μX=npσX=np(1−p)\mu_X = np \qquad \sigma_X = \sqrt{np(1-p)}

The shape depends on pp. The distribution is symmetric when p=0.5p = 0.5, skewed right when p<0.5p < 0.5, and skewed left when p>0.5p > 0.5. Because the variable is discrete, a single value has a positive probability, and P(X≤4)P(X \le 4) and P(X<4)P(X < 4) are different numbers. See the binomial probability guide for "at least" and "at most" calculations.

What a normal distribution is

A normal distribution is a continuous, unimodal, bell-shaped, symmetric curve. Two parameters identify it: the mean μ\mu, which locates the center, and the standard deviation σ\sigma, which sets the spread. A normal curve is symmetric for every choice of μ\mu and σ\sigma.

Probability is area under the curve, and the total area is 1. To find P(a<X<b)P(a < X < b), convert the boundaries to z-scores with z=x−μσz = \frac{x - \mu}{\sigma} and use the z-table, or use technology. The empirical rule gives quick estimates: about 68%, 95%, and 99.7% of values lie within 1, 2, and 3 standard deviations of the mean.

Because the variable is continuous, the probability of any single exact value is 0. Only intervals have probability, so P(X≤160)P(X \le 160) and P(X<160)P(X < 160) are the same number.

The differences side by side

FeatureBinomial distributionNormal distribution
Type of variableDiscrete countContinuous measurement
Possible values0,1,2,…,n0, 1, 2, \ldots, nAny real number
Parametersnn and ppμ\mu and σ\sigma
Meannpnpμ\mu
Standard deviationnp(1−p)\sqrt{np(1-p)}σ\sigma
ShapeSymmetric only when p=0.5p = 0.5, otherwise skewedAlways symmetric and bell-shaped
Probability of one exact valuePositive, from the formula0
Probability is found byAdding bar heightsFinding an area under the curve
≤\le versus <<Different answersSame answer
TI-84 commandsbinompdf, binomcdfnormalcdf, invNorm

In a binomial distribution the mean and standard deviation are both fixed by nn and pp. In a normal distribution they are two separate parameters, and either can be changed without touching the other.

When to use which

Use the binomial distribution when the variable counts how many times something happens in a fixed number of independent yes-or-no trials: the number of defective items in a batch of 20, the number of correct guesses on 10 multiple-choice questions, the number of voters in a sample of 50 who favor a measure.

Use the normal distribution when the variable is a continuous measurement that the problem says is normally or approximately normally distributed: heights, weights, test scores treated as continuous, measurement errors.

The two meet when nn is large. If np≥10np \ge 10 and n(1−p)≥10n(1-p) \ge 10, the binomial histogram is close to a normal curve with mean npnp and standard deviation np(1−p)\sqrt{np(1-p)}. That is why a sample proportion can be handled with a normal model in inference. Technology gives exact binomial probabilities, so the approximation is mainly a way to reason about shape, as explained in the normal approximation to the binomial.

The classic mix-up and how to avoid it

The first mix-up is using a normal calculation on a small binomial count. With n=20n = 20 and p=0.1p = 0.1, the expected number of successes is only 2, the distribution is strongly skewed right, and a symmetric curve does not fit. Use the binomial formula or binomcdf.

The second is carrying discrete habits into a normal problem, or the reverse. For a binomial variable, "fewer than 4" means X≤3X \le 3 and "at most 4" means X≤4X \le 4, and the two answers differ by P(X=4)P(X = 4). For a normal variable there is nothing to adjust, because a single point has probability 0.

The third is asking a normal model for the probability of one exact value and expecting a positive answer. A normal curve gives P(X=50)=0P(X = 50) = 0. If the variable is really a count, the binomial gives the right positive value.

A reliable first step is to ask what the variable is. A count of successes out of nn points to the binomial. A measurement described as approximately normal points to the normal. The broader distinction is in discrete vs continuous variables.

A binomial count

A basketball player makes 30% of her three-point attempts, and attempts are independent. She takes 20 attempts. Let X be the number she makes. Find the mean and standard deviation of X, the probability she makes exactly 6, and the probability she makes at most 4.

  1. Confirm the model: 20 fixed trials, make or miss, p=0.3p = 0.3 on each, independent. XX is binomial with n=20n = 20 and p=0.3p = 0.3.

  2. Mean: μX=np=20(0.3)=6\mu_X = np = 20(0.3) = 6 makes.

  3. Standard deviation: σX=np(1−p)=20(0.3)(0.7)=4.2≈2.049\sigma_X = \sqrt{np(1-p)} = \sqrt{20(0.3)(0.7)} = \sqrt{4.2} \approx 2.049 makes.

  4. Exactly 6: P(X=6)=(206)(0.3)6(0.7)14P(X = 6) = \binom{20}{6}(0.3)^6(0.7)^{14}. Here (206)=38760\binom{20}{6} = 38760, (0.3)6=0.000729(0.3)^6 = 0.000729, and (0.7)14≈0.006782(0.7)^{14} \approx 0.006782, so P(X=6)≈0.1916P(X = 6) \approx 0.1916. This is binompdf(20, 0.3, 6).

  5. At most 4: add the bars for 0 through 4, which is binomcdf(20, 0.3, 4): P(X≤4)≈0.2375P(X \le 4) \approx 0.2375.

  6. Note the discrete detail: 'fewer than 4' would be P(X≤3)P(X \le 3), a different and smaller number.

The mean is 6 makes and the standard deviation is about 2.049 makes. P(X=6)≈0.1916P(X = 6) \approx 0.1916 and P(X≤4)≈0.2375P(X \le 4) \approx 0.2375.

The same count through a normal curve

A fair coin is flipped 100 times and X is the number of heads. Find the exact binomial probability of at most 45 heads, then approximate it with a normal curve, and compare. Also compare what each model says about exactly 50 heads.

  1. Set up the binomial: n=100n = 100, p=0.5p = 0.5, so μX=100(0.5)=50\mu_X = 100(0.5) = 50 and σX=100(0.5)(0.5)=25=5\sigma_X = \sqrt{100(0.5)(0.5)} = \sqrt{25} = 5.

  2. Check the large counts: np=50np = 50 and n(1−p)=50n(1-p) = 50, both at least 10, so a normal curve with mean 50 and standard deviation 5 is a reasonable approximation.

  3. Exact binomial: binomcdf(100, 0.5, 45) gives P(X≤45)≈0.1841P(X \le 45) \approx 0.1841.

  4. Plain normal approximation: z=45−505=−1z = \frac{45 - 50}{5} = -1, and the area to the left of z=−1z = -1 is 0.15870.1587. That is about 2.5 percentage points below the exact value.

  5. Explain the gap. The binomial bar at 45 covers 44.5 to 45.5 on the number line, and cutting the curve at 45 leaves out half of it. Cutting at 45.5 instead gives z=45.5−505=−0.9z = \frac{45.5 - 50}{5} = -0.9, with area 0.18410.1841 to the left, which matches the exact value to four decimals.

  6. Exactly 50 heads: the binomial gives P(X=50)≈0.0796P(X = 50) \approx 0.0796. The normal curve gives 0 for any single value, because it is continuous.

The exact binomial probability is P(X≤45)≈0.1841P(X \le 45) \approx 0.1841. The plain normal approximation gives 0.15870.1587, and measuring to the edge of the bar at 45.5 gives 0.18410.1841. For exactly 50 heads the binomial gives about 0.07960.0796 and the normal model gives 0. The count is discrete, so the binomial is the exact model and the normal curve is an approximation to it.

Frequently asked questions

Is the binomial distribution a normal distribution?

No. A binomial distribution is discrete and can be skewed, and a normal distribution is continuous and always symmetric. When np≥10np \ge 10 and n(1−p)≥10n(1-p) \ge 10, a binomial distribution is close in shape to a normal curve, but it remains a separate distribution.

When can I use the normal distribution to approximate a binomial?

When both the expected number of successes npnp and the expected number of failures n(1−p)n(1-p) are at least 10. The approximating curve has mean npnp and standard deviation np(1−p)\sqrt{np(1-p)}. If technology is available, the exact binomial probability is the better answer.

Why is the probability of an exact value zero for a normal distribution but not for a binomial?

A normal variable is continuous, so probability is area, and a single point has no width and no area. A binomial variable is discrete, so each whole-number count has its own bar with a positive height.