Central Limit Theorem Calculator
The central limit theorem says that the average of a large enough sample behaves like a normal distribution, centered on the population mean with spread sigma / sqrt(n), whatever shape the population itself has. This calculator turns that promise into numbers: sampling probabilities, the sample mean that cuts off any target probability, confidence intervals, the sample size needed for a target precision, and a live simulation that shows the theorem happening.
One naming note before the math: despite the word limit, this page has nothing to do with the calculus limit, the value a function approaches as its input closes in on a point. The limit here is a limiting distribution as sample size grows. If your problem is about derivatives or function behavior, this is the wrong tool; if it is about averages of samples, it is exactly the right one.
Simulation mode uses its own built-in populations (uniform, exponential, normal) and ignores the mu and sigma fields, which belong to the other three modes.
Quick Answer
Key takeaways
- The sampling distribution of the mean is centered at mu and has spread sigma / sqrt(n). That spread, the standard error, is the only new quantity the theorem asks you to compute.
- Shape stops mattering as n grows: skewed, flat and bimodal populations all produce near-normal sample means, which is why one normal table serves every problem.
- Precision scales with the square root of n. Halving the standard error costs four times the sample, never two times.
- Probabilities about a mean use the standard error in the z-score. Using sigma instead is the classic error, and it inflates every probability toward the middle.
How to use the central limit theorem calculator
Every mode starts from the same two population values: the population mean mu and the population standard deviation sigma, entered at the top. Probability mode then takes a sample size and a target: the chance the sample mean falls below a value, above a value, or between two values. It returns the standard error, the z-scores, the probability as a decimal and a percent, the step-by-step working, and the 95% and 99% confidence intervals for the sample mean alongside.
Quantile mode runs the calculation in reverse: give it a probability p and it returns the sample mean value with that probability below it (or above it, if you switch the tail). Sample size mode solves for n given a target standard error, or a target confidence interval half-width at 95% or 99%. Simulation mode ignores mu and sigma and instead draws thousands of real random samples from a uniform, exponential or normal population you choose, reporting the mean and spread of the sample means against the theoretical values.
The z-score machinery in the first two modes is the same standard normal calculation used by the area to z-score calculator, with the standard error standing in for the standard deviation. For judging whether a finished study result clears a significance threshold, the statistical significance calculator takes over from where sampling distributions leave off, and the compound probability calculator covers the event-combination problems that sit next to this material in most courses.
The formulas, stated plainly
Standard error: SE = sigma / sqrt(n).
Z-score for a sample mean: z = (x-bar - mu) / SE.
Quantile: x-bar = mu + z x SE, with z from the inverse normal for your target probability.
Sample size: n = (sigma / SE)², always rounded up.
These are exactly the statements in the standard textbook treatment, for example OpenStax Statistics chapter 7: the sample means form their own normal distribution with the same mean as the original distribution and a standard deviation of sigma / sqrt(n), called the standard error of the mean. The theorem itself is the claim that this normal approximation arrives as n grows, for any population distribution with a finite mean and standard deviation.
The square root is the part intuition resists. Averaging n observations divides the variance by n, and standard deviation is the square root of variance, so the spread of the means shrinks by sqrt(n), not by n. Going from n = 35 to n = 140, four times the data, only halves the standard error from 1.1832 to 0.5916. Survey designers live inside that trade-off: the sample size mode exists because the last increment of precision is always the most expensive one.
Worked example, checked against this exact tool
The calculator is pre-filled with this case in Probability mode, and every number below is its own output.
| Step | Calculation | Result |
|---|---|---|
| Standard error | 7 / sqrt(35) | 1.1832 |
| Lower z-score | (68 - 70) / 1.1832 | -1.6903 |
| Upper z-score | (72 - 70) / 1.1832 | 1.6903 |
| Probability between 68 and 72 | Phi(1.6903) - Phi(-1.6903) | 90.9031% |
| 95% CI for the sample mean | 70 plus or minus 1.96 x 1.1832 | 67.6809 to 72.3191 |
| 99% CI for the sample mean | 70 plus or minus 2.576 x 1.1832 | 66.9520 to 73.0480 |
The other modes answer the natural follow-ups on the same population. In Quantile mode with n = 35, the value with 95% of sample means below it is 70 + 1.6449 x 1.1832 = 71.9462. In Sample size mode, a target standard error of 1 needs n = (7 / 1)² = 49. And in Simulation mode, drawing 1,000 samples of n = 30 from the flat uniform population produces a mean of means near 0.5 with a spread near the theoretical 0.2887 / sqrt(30) = 0.0527, the theorem made visible: a population with no bell shape at all, whose averages pile into one.
When the normal approximation can mislead you
The theorem is a limit result: it promises convergence as n grows, and stays silent about any particular finite n. Three situations deserve care. Heavily skewed populations, incomes and insurance claims are the classic cases, converge slowly, and a sample of 30 from a strongly skewed population can still produce a visibly skewed distribution of means. Small samples from a population that is itself close to normal are fine almost immediately, by contrast, because there is little shape to wash out.
The second trap is dependence. The theorem assumes independent draws; samples with clustering, time trends or repeated measurement of the same units violate that, and the true spread of the mean is larger than sigma / sqrt(n) suggests. The third is a population with no finite variance, such as the Cauchy distribution, where the theorem simply does not apply and sample means never settle down. For ordinary course and survey work none of these is likely to bite, but the simulation mode is the honest check: if the simulated spread of means disagrees with sigma / sqrt(n), believe the disagreement and ask why.
What competing calculators miss
The typical rival CLT tool computes a standard error and one probability, then stops. Working through the field for this page: one polished competitor adds a quantile mode and a visual explanation, but offers no sample-size solver and no confidence intervals beside the probability. Another returns probabilities with instant updating and a density display, but cannot run the calculation backwards to a cutoff value, and cannot answer the question practitioners ask most, which is how large a sample they need.
This page combines all four directions in one tool: probability (below, above and between, with 95% and 99% intervals attached), quantile, a sample-size solver that handles both target standard error and target interval half-width, and a live Monte Carlo simulation across three population shapes so the theorem can be watched rather than believed. The step-by-step working is shown in every mode, because a CLT answer you cannot audit teaches nothing.
One honest limit: the probability and quantile modes use the normal approximation throughout, with the standard Abramowitz and Stegun style CDF evaluation. For very small samples from a known-normal population, a t-distribution calculation would be more exact; this tool is built for the large-sample setting the theorem describes.
Add This Calculator to Your Website
Frequently Asked Questions
The standard error is the standard deviation of the sampling distribution of the mean: SE = sigma / sqrt(n). It measures how much sample means typically vary from the population mean. Because n sits under a square root, quadrupling the sample size only halves the standard error, which is why precision gets expensive.
The common rule of thumb is n = 30, but it is a rule of thumb, not a law. Roughly symmetric populations produce near-normal sample means at much smaller n, while heavily skewed populations can need n of 50 or more. The simulation mode on this page lets you watch the difference directly by running a skewed exponential population at n = 5 and then at n = 50.
Standard deviation, sigma, describes the spread of individual observations around the population mean. Standard error, sigma / sqrt(n), describes the spread of sample means around that same mean. Confusing the two is the most common central limit theorem mistake: probabilities for a sample mean must use the standard error, or they will be far too wide.
Use the Sample size mode. Enter sigma and either a target standard error, or a target confidence interval half-width with a confidence level. The tool inverts the standard error formula, n = (sigma / SE) squared, and rounds up, because rounding down would leave you with a slightly larger standard error than the one you asked for.
Yes, that is the entire point of the theorem. For a large enough sample, the distribution of the sample mean is approximately normal whatever the population shape: flat, skewed or bimodal. What changes with shape is how large is large enough, which is why the theorem is stated as a limit and why checking with the simulation is worthwhile for skewed data.
No, despite the shared word. A calculus limit describes what value a function approaches as its input approaches some point. The central limit theorem uses limit in the looser sense of limiting distribution: it describes what the distribution of sample means approaches as the sample size grows without bound. The two ideas live in different subjects and share nothing but the name.
Because they are two readings of the same sampling distribution. The 95% interval, mu plus or minus 1.96 standard errors, is the range that captures the middle 95% of all sample means. Showing it beside a tail or between-values probability helps you see whether your target value is ordinary sampling variation or something genuinely unusual.
References
- OpenStax, Statistics, "The Central Limit Theorem for Sample Means (Averages)": the sampling distribution statement, the z-score for a sample mean, and the standard error of the mean. OpenStax: The Central Limit Theorem for Sample Means. Accessed October 2026.
- OpenStax, Introductory Statistics, "The Central Limit Theorem for Sample Means (Averages)": the same chapter in the introductory text, including worked sampling distribution examples. OpenStax: Introductory Statistics, Central Limit Theorem. Accessed October 2026.
- Statistics LibreTexts, "The Central Limit Theorem for Sample Means (Averages)": the OpenStax chapter with additional exercises on standard error and sample size. Statistics LibreTexts: The Central Limit Theorem. Accessed October 2026.