Normal Distribution Calculator

What the normal distribution describes

The normal distribution is a symmetric bell-shaped model for continuous measurements. It has a mean, written as μ, and a standard deviation, written as σ. Values near the mean have more density than values far away. The complete curve has area 1, or 100 percent. Probability refers to an interval, such as below a cutoff or between two bounds. It can model some measurement errors, repeated readings, and aggregated scores. It is not proof that every data set is normal. Inspect skew, outliers, and multiple peaks before relying on it. NIST gives the density formula f(x) = exp(-(x - μ)^2 / (2σ^2)) / (σ sqrt(2π)) in its normal distribution reference.

There are two related but different ideas: density and cumulative probability. Density describes the height of the curve at a location, while cumulative probability describes the area collected up to that location. A high density does not mean a single exact value has a large probability in a continuous model. The interval width matters. A narrow interval near the mean may contain less area than a wider interval farther away.

Parameters describe a population model, while statistics describe a sample. A sample mean and sample standard deviation are estimates of μ and σ, not perfect truths. If the sample is small, the fitted curve can be unstable. If the sample includes a mixture of groups, one bell curve may hide meaningful differences. State whether the numbers came from a known population, a sample, or a fitted process when sharing a result.

Turning a value into a z score

The z score puts a raw value on the standard normal scale. Use z = (x - μ) / σ. Subtract the mean, then divide by the standard deviation. A positive z is above the center, a negative z is below it, and zero is exactly the mean. Example 1: mean 100 and standard deviation 15, with x = 130, gives z = (130 - 100) / 15 = 2. The left area is about 97.7 percent and the upper tail about 2.3 percent. Example 2: mean 42 minutes and standard deviation 6, with x = 30, gives z = -2 and a left area near 2.3 percent.

The sign of a z score is useful for direction, while its magnitude is useful for distance. Values with z = 1.5 and z = -1.5 have matching two-sided extremeness because the curve is symmetric, but they answer different one-sided questions. A left-tail request should preserve the sign. For a two-sided unusualness question, calculate both tails or use the area outside the matching central interval.

A table may show area from zero to z instead of area from negative infinity to z. For z = 1, a zero-to-z table gives about 0.3413, while the left cumulative area is 0.8413. Add 0.5 for a positive z when using that table. For a negative z, use symmetry. Always read the table heading before copying a value into a calculation.

Reading tails and intervals

The left tail is P(X ≤ x), the area to the left of a value. The right tail is P(X ≥ x), which is 1 minus the left area. An interval uses subtraction: upper cumulative area minus lower cumulative area. For mean 50 and standard deviation 10, the range 40 to 60 has z values -1 and 1. Its area is about 0.8413 - 0.1587 = 0.6826, or 68.26 percent. “At most” means left tail. “At least” means right tail. NIST standard normal tables show the same area convention at this reference table.

The 68 95 99.7 rule

A normal model puts about 68 percent of values within 1 standard deviation, 95 percent within 2, and 99.7 percent within 3. More precise values are 68.27 percent, 95.45 percent, and 99.73 percent. If mean is 500 and standard deviation is 40, the one standard deviation interval is 460 to 540, the two standard deviation interval is 420 to 580, and the three standard deviation interval is 380 to 620. These are long-run model shares, not guarantees for a small sample. The rule is not a repair for skewed data. Check the shape and source population first. Keep all inputs in matching units.

Percentiles and cutoffs

A percentile is a cumulative position. The 50th percentile is the mean in a normal model. The 90th percentile uses z about 1.282. With mean 100 and standard deviation 15, its cutoff is 100 + 1.282 x 15, or about 119.2. The lowest 5 percent uses z near -1.645. With mean 70 and standard deviation 8, that cutoff is about 56.8. A percentile is not a percent correct. It describes the fraction of modeled observations at or below a value. Different populations or fitted parameters can produce different percentiles.

Use the estimate with care

Enter a positive standard deviation and matching units. The widget returns a z score, left tail, right tail, and probability between two bounds. Reversed bounds are ordered before subtraction. A zero spread cannot define a normal curve, so the tool warns about it. The calculation runs in the browser using the standard normal cumulative function and does not upload entries. If mean and standard deviation came from a sample, parameter uncertainty still matters. NIST’s normal model overview covers parameters, density, and cumulative probability.

To choose the right input, define the random variable before entering numbers. If X is a measured temperature, x must use the same temperature scale as μ and σ. A standard deviation is not a margin that every observation must fit inside. The curve assigns progressively smaller areas to more distant values, but it still permits values beyond 3 standard deviations. That distinction matters when someone treats the 99.7 percent rule as a physical limit.

Rounding deserves attention. A z score shown as 2.00 can come from a value slightly below or above 2 standard deviations. The probability should be computed from the unrounded z score. For a displayed result, two decimal places are usually readable, but a cutoff used for a rule may need more precision. Keep the raw inputs and the displayed settings in your notes.

The calculator is also useful for comparing measurements with different raw scales. A score of 130 in one test and a delivery time of 30 minutes are not directly comparable, but their z scores can be compared when each model is credible. Both examples above are 2 standard deviations from their own means, even though their units and contexts differ. This is the main practical value of standardization.

Finally, probability is not the same as frequency in a small group. If a model says 2.3 percent, a group of 100 may contain zero, one, two, or more observations in that tail. Larger samples tend to stabilize proportions, but randomness remains. Report the model, its parameters, and the event definition along with the percentage.

Choosing a normal model

A normal model is most useful when the measurement is continuous, the observations are reasonably independent, and the distribution is close to symmetric. A histogram, dot plot, or normal quantile plot can reveal whether that description is sensible. Mild skew may be acceptable for a rough estimate near the center, but tail estimates are much more sensitive to shape. A few extreme observations can change the standard deviation and therefore move every reported z score.

Separate the question of fit from the question of calculation. Once μ and σ are supplied, the formulas are deterministic. That does not make the resulting probability trustworthy if the parameters came from a biased sample or if the process changes over time. Check the collection method, units, date range, and subgroup definition before treating a percentage as evidence.

Transformations can sometimes make a skewed measurement more symmetric, but the transformed result must be interpreted on its transformed scale. For counts, bounded percentages, waiting times, and financial amounts, another distribution may be more natural. If the main goal is a percentile or tail decision, compare the normal estimate with an empirical percentile from the observed data.

Use the calculator to show the model assumption openly. A clear report names the variable, mean, standard deviation, event, and direction of the tail. That short description lets another reader reproduce the result and decide whether the bell curve is appropriate.

Worked planning examples

Suppose a filling machine produces containers with a mean of 500 grams and a standard deviation of 8 grams. The chance that a container is under 492 grams uses z = (492 - 500) / 8 = -1. The left tail is about 15.87 percent. The chance between 492 and 508 grams is the area from -1 to 1, about 68.27 percent. These values describe the modeled production process, not a promise about the next container.

For a service time with mean 12 minutes and standard deviation 3 minutes, a 15 minute observation has z = 1. The left cumulative probability is about 84.13 percent, so the right tail is about 15.87 percent. If a manager needs the slowest 10 percent cutoff, use z about 1.282 and calculate 12 + 1.282 x 3, which is about 15.85 minutes. The cutoff is a planning threshold, not a claim that exactly 10 of every 100 calls will exceed it.

These examples also show why the direction matters. Under a lower specification uses a left tail, while exceeding an upper specification uses a right tail. A central tolerance uses two cumulative values and subtraction. Write the event in words before entering the numbers so the calculator output answers the intended question.

When a result is close to a policy boundary, retain extra decimal places internally and round only the displayed answer. A value of 15.85 minutes may be reported as 15.9 minutes for communication, while the unrounded value should support the actual decision. Include the parameters and units whenever the result is shared outside the calculator.