What Is the Difference Between a Normal and Binomial Distribution?
The binomial distribution is a discrete distribution for the number of successes in a fixed number of Bernoulli trials. The normal distribution is a continuous, symmetric, bell-shaped distribution described by a mean and standard deviation. A binomial distribution can sometimes be approximated by a normal distribution when its expected numbers of successes and failures are both sufficiently large.
The fastest way to choose between them is to identify the random variable. If you are counting successes out of a fixed number of trials, a binomial model may fit. If you are modeling a continuous measurement that is reasonably represented by a symmetric bell-shaped distribution, a normal model may fit. Sample size alone does not make that decision.
- Count of successes out of n trials: check the binomial assumptions.
- Continuous measurement: consider whether a normal model is scientifically and empirically reasonable.
- Approximating a binomial: examine both np and n(1-p), not n by itself.
- Sampling without replacement: independence may fail, so a hypergeometric model can be more appropriate.
Normal vs Binomial Distribution at a Glance
| Feature | Binomial Distribution | Normal Distribution |
|---|---|---|
| Variable type | Discrete count | Continuous numerical variable |
| Notation | X ~ Binomial(n, p) | X ~ N(μ, σ²) |
| Parameters | n = trials, p = success probability | μ = mean, σ = standard deviation |
| Possible values | 0, 1, 2, ..., n | Any real number mathematically |
| Probability function | Probability mass function (PMF) | Probability density function (PDF) |
| Point probability | P(X = k) can be positive | P(X = x) = 0 for one exact value |
| Mean | np | μ |
| Variance | np(1-p) | σ² |
| Shape | Depends on n and p; can be skewed or fairly symmetric | Always symmetric and bell-shaped |
| Typical use | Number of successes in repeated Bernoulli trials | Continuous measurements or other variables modeled with a normal distribution |
| Relationship | Can sometimes be approximated by a normal distribution | Can serve as an approximation to binomial under suitable conditions |
Discrete bars versus a continuous density curve
A count such as 4 or 5 is possible. A count such as 4.7 is not.
Probability is assigned to intervals. The height of the PDF at one point is a density, not a point probability.
What Is a Binomial Distribution?
Here, n is the number of trials, p is the probability of success on one trial and X is the number of successes. In this context, p is a success probability. It is not the p-value used in hypothesis testing.
Binomial assumptions
A classical binomial model needs more than a yes-or-no outcome. Check all five conditions:
- The number of trials n is fixed before observing the outcomes.
- Each trial can be classified into two categories for the event of interest: success or not success.
- The success probability p is the same for every trial.
- The trials are independent, or the model treats them as sufficiently independent for the purpose at hand.
- The random variable counts the total number of successes.
The underlying physical outcome does not need to be literally binary. When rolling a six-sided die, for example, you can define success as "roll a 6" and failure as "roll anything else." The event has two categories even though the die has six faces.
k = number of successes
C(n,k) = "n choose k"
p = success probability
The binomial coefficient C(n,k) counts how many arrangements contain exactly k successes among n trials. Each arrangement has the same probability when the assumptions above hold.
The mean np is a long-run expected count across repeated versions of the full n-trial experiment. It does not mean one experiment must produce exactly np successes. Penn State's probability course gives the same mean and variance formulas and shows how the shape changes with n and p.
Exact binomial probability: 6 heads in 10 fair coin tosses
Let X = number of heads. Then X ~ Binomial(10, 0.5).
For exactly 6 heads, use k = 6: P(X = 6) = C(10,6)(0.5)6(0.5)4.
C(10,6) = 210, so P(X = 6) = 210 / 1024 = 0.205078125.
Answer: P(X = 6) ≈ 0.2051, or about 20.51%.
If you sample from a finite population without replacement, the trials are not strictly independent because the composition of the population changes after each draw. In that setting, a hypergeometric model may be more appropriate than a binomial model.
What Is a Normal Distribution?
The mean μ sets the center of the curve. The standard deviation σ controls its spread, while σ² is the variance. Every theoretical normal distribution is symmetric and unimodal. Its mean, median and mode coincide at μ.
The mathematical support extends over all real numbers. A real variable may still have practical boundaries. Human height, for example, cannot be negative, even if a normal model with sensible parameters assigns an effectively negligible amount of density to impossible negative values. That is one reason normality should be treated as a model, not a claim that nature follows a formula perfectly.
μ = center
σ = spread
σ² = variance
For a continuous random variable, the probability of one exact point is zero. Probabilities come from areas over intervals. Saying f(70) = 0.08 does not mean P(X = 70) = 0.08. The first quantity is density; the second is a probability and equals zero for a single point in a continuous model.
Normal probability within one standard deviation
Suppose a continuous measurement is reasonably modeled as X ~ N(70, 5²). What is the probability that X falls between 65 and 75?
Standardize 65: z = (65 - 70) / 5 = -1.
Standardize 75: z = (75 - 70) / 5 = 1.
The standard normal area between z = -1 and z = 1 is approximately 0.6827.
Answer: P(65 ≤ X ≤ 75) ≈ 0.6827, or about 68.27%.
For more detail on standardizing values, see the z-score guide and the normal distribution calculator.
Key Differences Between Binomial and Normal Distributions
1. Discrete versus continuous
A binomial random variable is a count. If X records the number of successful trials, values such as 0, 1, 2 and 3 make sense; 2.4 successes does not. A normal random variable is continuous and can take any value along an interval, at least within the mathematical model.
2. Probability mass versus probability density
The binomial PMF gives the probability of an exact count, so P(X = 4) can be positive. The normal PDF gives density. Probabilities are obtained by integrating the density over an interval. This distinction matters when reading graphs: a binomial bar has probability mass, while the height of a normal curve at one point is not itself a probability.
3. Parameters and shape
Changing n or p can alter the binomial's center, spread and skewness. When p is near 0.5 and n is suitable, the shape may look fairly symmetric. When p is near 0 or 1, it can be noticeably skewed. A normal distribution is always symmetric by definition; μ shifts the center and σ stretches or compresses the curve.
4. Bounded versus unbounded support
A binomial count is bounded between 0 and n. The normal distribution has mathematical support from negative infinity to positive infinity. That difference remains true even when the two distributions have nearly the same visual shape over the region where most probability lies.
5. Different scientific questions
Choosing a distribution is not mainly a curve-matching exercise. A binomial model describes repeated Bernoulli trials and counts the resulting successes. A normal model describes a continuous random variable with a particular symmetric density. Two datasets can have similar-looking histograms and still arise from different probability mechanisms.
When Should You Use Binomial or Normal?
Should I Use Binomial or Normal?
Use a binomial distribution when
- You have a fixed number of trials.
- Each trial is classified as success or not success for the event being studied.
- The success probability remains constant.
- The trials are independent enough for the model.
- Your random variable is the number of successes.
Use a normal distribution when
- The modeled variable is continuous.
- A symmetric bell-shaped model is reasonable for the scientific process or analytical purpose.
- Or the object being modeled is a sampling distribution that is approximately normal under justified conditions.
"Use normal when n is large" and "use binomial when n is small" are not reliable model-selection rules. A binomial model can be exact for n = 5 or n = 50,000 when its assumptions fit. A normal distribution is a continuous probability model regardless of sample size.
When Can a Binomial Distribution Be Approximated by a Normal Distribution?
A binomial random variable can sometimes be approximated by a normal random variable because the binomial count is the sum of Bernoulli random variables. Central-limit-theorem reasoning explains why that sum can become approximately normal under suitable conditions. The approximation does not turn the discrete distribution into a continuous one; it replaces one probability calculation with a close continuous approximation.
If X ~ Binomial(n,p), the matching normal approximation uses the same mean and variance:
μ = np
σ = √[np(1-p)]
Approximation quality depends on both n and p. Classroom rules often require np and n(1-p) to be at least 5 or 10, while some courses use stricter thresholds. These are rules of thumb rather than a universal theorem. They work because np and n(1-p) are the expected numbers of successes and failures. If either expected count is very small, the binomial distribution can remain strongly skewed.
| Situation | Approximation tendency | Why |
|---|---|---|
| Moderate/large n, p near 0.5 | Usually better | The binomial is more symmetric and both expected counts are typically substantial. |
| Small n | Often worse | There may be too little probability mass for a smooth curve to track the discrete bars closely. |
| p near 0 or 1 | Often worse | The distribution can be strongly skewed. |
| Large n but extremely small p | Can still be poor | A large n alone does not guarantee that np is large. |
| Both np and n(1-p) comfortably large | Usually better | Expected successes and failures are both large enough for the shape to be less skewed. |
| Tail probability requiring high precision | Check exact result | Small approximation errors can matter more in the tails. |
Take n = 100 and p = 0.001. Then np = 0.1 and n(1-p) = 99.9. One expected count is tiny, so the binomial remains highly concentrated near zero. Calling it "approximately normal" simply because n is 100 would ignore the shape created by p.
Continuity correction
A binomial model places probability at integers. A normal approximation spreads probability continuously. A continuity correction shifts an integer boundary by 0.5 so that a normal area lines up more closely with the discrete bars it is replacing.
left edge
binomial count
right edge
| Binomial probability | Normal approximation boundary | Reason |
|---|---|---|
| P(X ≤ 10) | P(Y ≤ 10.5) | Include the full probability bar centered at 10. |
| P(X ≥ 10) | P(Y ≥ 9.5) | Start at the left edge of the bar centered at 10. |
| P(X = 10) | P(9.5 ≤ Y ≤ 10.5) | Replace the single discrete bar with its matching continuous interval. |
For a dedicated treatment, see the site's normal approximation guide and continuity correction calculator.
Worked Example: Normal Approximation to a Binomial Probability
Approximate P(X ≤ 45) when X ~ Binomial(100, 0.40)
Here n = 100 and p = 0.40. The expected successes and failures are 40 and 60, so this is a relatively favorable shape for a normal approximation.
Match the mean: μ = np = 100(0.40) = 40.
Match the standard deviation: σ = √[np(1-p)] = √[100(0.40)(0.60)] = √24 ≈ 4.899.
Apply continuity correction: P(X ≤ 45) becomes approximately P(Y ≤ 45.5).
Standardize: z = (45.5 - 40) / 4.899 ≈ 1.123.
Read the normal CDF: Φ(1.123) ≈ 0.8692.
Normal approximation: about 0.8692. The exact binomial probability is about 0.8689, a difference of roughly 0.0003 in this example.
The close result here should not be generalized to every binomial problem. With more extreme p values, smaller expected counts or sensitive tail probabilities, the gap can be larger. When exact binomial calculation is easy, it remains the reference calculation.
Normal vs Binomial Examples
Coin tosses
Number of heads in 20 independent tosses with constant p = 0.5: binomial.
Defective units
Count of defective parts in a fixed batch under a stable, independent defect model: binomial candidate.
Measurement error
A continuous measurement error that is approximately symmetric and bell-shaped may be modeled with a normal distribution.
Lab measurement
A continuous assay reading may sometimes be modeled as normal after checking scientific context and distributional fit.
Correct answers
Number correct out of a fixed set of independent binary-scored items with constant p under a simplified model: binomial.
Sampling distribution
A statistic such as a sample mean can have an approximately normal sampling distribution under appropriate conditions.
For topic-specific examples, see binomial distribution examples and normal distribution examples.
Common Normal vs Binomial Mistakes
| Mistake | Why it fails | Better rule |
|---|---|---|
| "Binomial is for small samples." | Sample size does not define the binomial model. | Use binomial for counts of successes when the Bernoulli-trial assumptions fit. |
| "Normal is for large samples." | The normal distribution is a continuous probability model, not a sample-size category. | Use it when the variable or sampling distribution is reasonably modeled as normal. |
| Every yes/no dataset is binomial. | Changing p or dependent trials can violate the model. | Check fixed n, constant p, independence and the success-count variable. |
| All binomial distributions are bell-shaped. | Extreme p values can create substantial skewness. | Shape depends on both n and p. |
| Every bell-shaped dataset is normal. | Many non-normal distributions can look roughly bell-shaped in finite samples. | Treat normality as a model that must be justified for the analysis. |
| n ≥ 30 guarantees normal approximation. | A large n can still pair with an extreme p and tiny np. | Examine np and n(1-p), shape, tail sensitivity and required accuracy. |
| Forget continuity correction. | A continuous area can be misaligned with discrete integer bars. | Shift relevant integer boundaries by 0.5 when using the approximation. |
| PDF height equals probability. | For a continuous variable, point probability is zero. | Use area over an interval for normal probabilities. |
| Normal approximation makes binomial normal. | The original binomial distribution stays discrete. | Call it an approximation, not an identity. |
Frequently Asked Questions
A binomial distribution models a discrete count of successes in a fixed number of Bernoulli trials, while a normal distribution models a continuous variable with a symmetric bell-shaped density. Binomial parameters are n and p. Normal parameters are μ and σ.
It is discrete. A binomial random variable counts successes, so its possible values are integers from 0 through n. Values such as 4.7 successes are impossible.
It is continuous. The theoretical normal distribution has support across all real numbers, and probabilities correspond to areas over intervals under its density curve.
No finite binomial distribution literally becomes normal. It remains a discrete distribution on the integers 0 through n. Under suitable conditions, however, its probabilities can be approximated closely by a normal distribution with mean np and variance np(1-p).
Use it when the binomial shape is sufficiently well behaved, commonly assessed by checking that both np and n(1-p) are large enough under the convention used in your course or application. The required accuracy, skewness and tail probability also matter.
Not by itself. Approximation quality depends strongly on p as well as n. For example, n = 100 with p = 0.001 gives np = 0.1, so the binomial remains highly concentrated near zero rather than looking symmetric and normal.
Continuity correction adjusts a discrete binomial boundary by 0.5 when it is replaced with a continuous normal area. For example, P(X ≤ 10) is approximated with P(Y ≤ 10.5), and P(X = 10) with P(9.5 ≤ Y ≤ 10.5).
No. The binomial model allows any success probability from 0 to 1. The value of p affects the center, spread, skewness and the quality of a possible normal approximation.
A Bernoulli random variable describes one trial with success coded as 1 and failure as 0. A binomial random variable counts successes across n independent Bernoulli trials with the same success probability p.
- Start with the variable you are modeling, not the sample size.
- Binomial means a discrete success count under specific Bernoulli-trial assumptions.
- Normal means a continuous symmetric bell-shaped model with parameters μ and σ.
- A normal approximation to binomial is optional and approximate; it does not change the original distribution.
- When approximation is used, match μ = np and σ = √[np(1-p)] and consider continuity correction.
Sources and References
The mathematical definitions, formulas and approximation discussion were checked against university and government statistical references.
- Penn State STAT 414: The Binomial Distribution. Conditions, PMF, shape, mean and variance. online.stat.psu.edu
- NIST/SEMATECH Engineering Statistics Handbook: Normal Distribution. PDF, parameters and standard normal distribution. itl.nist.gov
- Penn State STAT 414: Approximations for Discrete Distributions. Normal approximation to the binomial and the connection to the Central Limit Theorem. online.stat.psu.edu
- Penn State STAT 500: Sampling Distributions. Discussion of changing rule-of-thumb thresholds based on expected successes and failures. online.stat.psu.edu