Z-Score Calculator

What a z-score actually measures

A z-score answers a simple question with a precise number: relative to the spread of the whole dataset, how unusual is this value? The mean alone cannot answer it. A height of 180 cm is unremarkable for adult men in the Netherlands but notable in Japan, and the raw number does not tell you which. What matters is not the distance from the average but that distance measured in units of the distribution's own spread.

That unit is the standard deviation. A z-score of exactly 0 means the value equals the mean. A z-score of 1 means one standard deviation above the mean; a z-score of -2 means two standard deviations below it. Because the unit is the data's own spread, z-scores are dimensionless. You can z-score heights in centimeters, salaries in dollars, and sprint times in seconds, and the resulting numbers are directly comparable even though the underlying quantities are not.

Converting to z-scores puts every normal distribution on the same ruler: the standard normal, with mean 0 and standard deviation 1. On that ruler you can look up how rare a value is and how much of the distribution sits above or below it.

The z-score formula, variable by variable

The formula is z = (x - mu) / sigma. Each symbol does one job, and each one is a place where a mistake can silently wreck the answer.

x is the raw score, the single value you are evaluating: one exam result, one measurement, one observation. It must come from the same distribution as the mean and standard deviation. Z-scoring a measurement against the wrong population's parameters is the most common misuse of the formula.

mu is the population mean, the average of the distribution the value belongs to. If you only know a sample mean, the z-score you compute is an approximation, and with small samples it can be badly off. In that situation, statistics uses a t-score with a t-distribution instead, because the extra uncertainty of an estimated mean widens the tails.

sigma is the population standard deviation, the typical distance between values and the mean. Dividing by sigma is what converts distance from the mean into number of standard deviations. A small sigma makes the same raw distance produce a large z-score; a large sigma dilutes it. Two exam scores both 10 points above their class means mean completely different things if one class has sigma of 5 and the other 20.

Sign is information, not error. A negative z-score is a perfectly good result meaning below the mean; it is not a mistake to be corrected. When you square z-scores or compare them to thresholds, keep the sign until the last step.

Worked examples, step by step

Example 1: above the mean. A student scores 85 on an exam where the class mean is 70 and the standard deviation is 10.

Step 1: subtract the mean. 85 - 70 = 15. Step 2: divide by the standard deviation. 15 / 10 = 1.5. The z-score is 1.5, so the score sits one and a half standard deviations above the class average. Under a normal curve that is roughly the 93rd percentile: about 93 percent of the class scored lower.

Example 2: below the mean. Another student scores 62 on the same exam. z = (62 - 70) / 10 = -8 / 10 = -0.8. The score is eight tenths of a standard deviation below the average, around the 21st percentile. The negative sign carries the below-average information; the magnitude 0.8 says how far.

Example 3: comparing across different tests. A student scores 1300 on an SAT where the mean is 1000 and the standard deviation is 200, giving z = (1300 - 1000) / 200 = 1.5. On an ACT-style test with mean 21 and standard deviation 5, the same student scores 28.5, giving z = (28.5 - 21) / 5 = 1.5. Identical z-scores mean identical relative standing: the performances are equally strong even though the raw numbers live on different scales. This is the everyday superpower of z-scores, and it is exactly how test agencies report percentile ranks.

In every case the steps are the same: subtract, divide, interpret the sign, then interpret the magnitude against a percentile table or a decision threshold.

Interpreting z-scores: from z to percentile

When the data follow a normal distribution, each z-score corresponds to a fixed percentile. These values come from the standard normal curve and are worth knowing in bands rather than to many decimals. The table below covers the values that come up constantly in practice, from quality control thresholds to the cutoffs behind confidence intervals.

Two cautions when reading this table. First, the percentiles are exact only for a true normal distribution; for skewed or heavy-tailed data the real shares can differ substantially. Second, in small samples a single unusual point can distort the mean and standard deviation themselves, so a large z-score in a dataset of ten values deserves less confidence than the same z-score in a dataset of ten thousand. The table describes the population, not your particular sample's quirks.

• Between z = -1 and z = 1 sits about 68 percent of a normal distribution.
• Between z = -2 and z = 2 sits about 95 percent, the famous empirical rule.
• Between z = -3 and z = 3 sits about 99.7 percent, so |z| above 3 happens roughly 3 times in 1,000.
• The z = 1.96 line is where the conventional 95 percent confidence interval comes from.
• Values beyond |z| = 3 are commonly labeled outliers in quality control and data cleaning.

Where z-scores are used in the real world

Education and testing. Standardized tests are scaled so that results can be compared across years and forms. Your reported score maps to a percentile, and that percentile is computed through the same z-score machinery: subtract the cohort mean, divide by the cohort standard deviation, look up the percentile. A phrase like above the 84th percentile is the friendly face of a z-score at or above 1.

Medicine and health. Growth charts for children are built on z-scores against age-matched reference populations. A child at height z = -1.5 is smaller than about 93 percent of peers of the same age and sex. Clinicians track how a child's z-score moves over time, because a stable negative z-score can be normal for that child while a sharp drop across visits is a signal worth investigating.

Manufacturing and quality control. Process capability analysis measures how many standard deviations fit between the process mean and the specification limits. Six Sigma methods are named after the goal of pushing defects out past six standard deviations on each side. Control charts flag any measurement beyond roughly three standard deviations from the target, which is the |z| > 3 line from the table above.

Finance and investing. Daily returns are routinely standardized before being compared. A trading day whose return is z = -4 against its own history is a shock day no matter what the market's average volatility is. Standardizing returns also lets risk models compare assets with wildly different volatilities on one scale.

Sports analytics. A basketball player scoring 30 points per game matters only relative to league scoring context. Analysts z-score player statistics against league-year means and standard deviations, so a 28-point season in a low-scoring era can score higher than a 32-point season in a high-scoring one. The same logic drives era-adjusted metrics and draft combine comparisons.

Assumptions and limitations you should know

The percentile interpretation rests on one assumption: the distribution is approximately normal. If your data are skewed, bimodal, or heavy-tailed, the z-to-percentile mapping from the table above does not hold. A z-score of 2 in a strongly right-skewed distribution like household income does not mean the 97.7th percentile; it means two standard deviations above the mean, full stop. Check the shape of your data before quoting percentiles.

The formula also assumes you know the population mean and standard deviation. In practice you often estimate them from a sample. With large samples the estimate is good and the z-score is fine. With small samples (roughly under 30) the extra uncertainty matters, and the t-distribution is the correct tool; z-scores computed from sample statistics tend to overstate how extreme a value is.

Z-scores are sensitive to outliers in a circular way. The mean and standard deviation are themselves distorted by extreme values, so one wild data point can inflate sigma enough to shrink every other z-score and hide real variation. Robust alternatives, such as standardizing by the median and median absolute deviation, exist for such datasets.

Finally, a z-score describes where a value sits, not why. It says nothing about causation, importance, or whether the value is an error. In data cleaning, |z| > 3 values are candidates for investigation, not automatic deletions: some of the most interesting observations in any dataset are exactly its most extreme z-scores.

Using this calculator well

Enter the raw score, the mean, and the standard deviation of the population the score belongs to. Be consistent about which population you are using: z-scoring your exam grade against your class is a different question from z-scoring it against all test takers nationwide, and both can be legitimate. The tool applies z = (x - mu) / sigma directly and reports the signed result.

Read the output in two parts. The sign tells direction: positive means above the mean, negative means below. The magnitude tells rarity under a normal curve: |z| under 1 is ordinary, between 1 and 2 is somewhat notable, between 2 and 3 is genuinely unusual, and beyond 3 is rare enough that you should double-check the inputs before celebrating or panicking.

To run the calculation in reverse, multiply and add: x = mu + z * sigma recovers a raw score from a z-score. That is how test designers set cut scores, for example choosing the raw score that corresponds to z = 1.5 on a given scale.

Disclaimer: this calculator is an educational tool for statistical estimation. It does not account for data quality, sampling error beyond what the inputs provide, or the specific distribution of your data. Verify results independently before using them for academic, medical, financial, or professional decisions.

• Use population parameters (mu and sigma) when you have them; treat sample-based z-scores as approximations.
• Quote percentiles only when the distribution is plausibly normal.
• Keep the sign of the z-score through any downstream calculations.
• For small samples, use a t-score or a nonparametric method instead.
• Treat extreme z-scores as questions to investigate, not verdicts to record.