Z-Score Calculator
Convert any raw value into a z-score — the number of standard deviations it sits above or below the mean of its distribution. Used everywhere from standardised test reporting and quality control to risk modelling and machine-learning feature scaling.
About this calculator
A z-score (also called a standard score) expresses where a value sits within a distribution by re-scaling it relative to the mean and standard deviation. The formula is: z = (X − μ) / σ, where X is the raw value, μ is the mean of the distribution (population mean if known, sample mean otherwise), and σ is the standard deviation. The result is dimensionless: a z-score of +1 means the value is one standard deviation above the mean; −2 means two standard deviations below. Because z-scores strip away the original units, they let you directly compare observations from completely different distributions — a student's math test score and verbal test score can be compared on a common scale even if the two tests have different means and spreads. For a normally distributed variable, z-scores map cleanly onto cumulative probabilities via the standard normal CDF: z = 1.0 corresponds to ~84.1th percentile, z = 1.645 to the 95th, z = 1.96 to the 97.5th, and z = 2.576 to the 99.5th. This is why z-scores are the workhorse of hypothesis testing and confidence intervals. Edge cases: σ must be > 0; if σ = 0 the formula divides by zero and the value has no meaningful z-score (every observation equals the mean). For small samples where σ is estimated and unknown, the t-score is technically more appropriate than the z-score because it accounts for additional uncertainty (the difference vanishes as n grows). The percentile interpretation assumes the underlying distribution really is normal — a z-score of 2 in a heavily skewed distribution does not correspond to the same percentile as in a normal distribution, so always verify the distributional assumption before using z-tables for probabilities.
How to use
Example 1 — Test score percentile. A student scores 85 on an exam where the class mean is 75 and standard deviation is 10. Enter Value = 85, Mean = 75, Standard Deviation = 10. z = (85 − 75) / 10 = 1.0. ✓ A z-score of 1.0 corresponds to roughly the 84.1st percentile of the standard normal — the student outperformed about 84% of classmates (assuming scores are normally distributed). Example 2 — Below-mean value. A patient's blood pressure reading is 110 mmHg, where the population mean for adults their age is 120 with a standard deviation of 12. Enter Value = 110, Mean = 120, Standard Deviation = 12. z = (110 − 120) / 12 = −0.833. ✓ The patient is about 0.83 standard deviations below the population mean — well within the normal range (roughly the 20th percentile), nothing remarkable. A reading of 160 (z = +3.33) would be in the top 0.05%, and clinically very high.
Frequently asked questions
What does a z-score actually tell me?
A z-score tells you exactly how many standard deviations a value lies from the mean of its distribution, and in which direction. Positive z-scores are above the mean, negative are below, and z = 0 is exactly at the mean. The interpretive power comes from normalisation: a z-score of +2 means the value is unusually high in any distribution, regardless of original units. For a normal distribution, you can also translate z directly into percentiles using a z-table or normal CDF: z = ±1 covers about 68% of the data, ±2 covers about 95%, ±3 covers about 99.7% (the 68-95-99.7 rule). For non-normal distributions the deviation interpretation still holds, but the percentile mapping does not.
How do I use a z-score to find probability or percentile?
Look up the z-score in a standard normal table (z-table) or compute Φ(z), the standard normal CDF, in software. Φ(z) gives the cumulative probability P(Z ≤ z) — equivalently, the percentile rank of the value. Common landmarks: Φ(1.00) ≈ 0.8413, Φ(1.645) ≈ 0.9500, Φ(1.96) ≈ 0.9750, Φ(2.576) ≈ 0.9950. For a right-tail probability use 1 − Φ(z); for a two-tailed probability use 2·(1 − Φ(|z|)). In Excel: NORM.S.DIST(z, TRUE) returns Φ(z). All of this assumes the underlying variable is approximately normally distributed; for heavily skewed data the same z does not correspond to the same probability, and you need a different distribution or an empirical CDF.
When should I use a z-score vs a t-score?
Use a z-score when the population standard deviation σ is genuinely known (rare in practice — usually only for well-characterised processes like standardised tests or established physical measurements) or when the sample size is large enough (typically n ≥ 30) that the sample standard deviation s is a near-perfect estimate of σ. Use a t-score when σ is unknown and you are estimating it from the same small sample — the t-distribution has heavier tails to account for the extra uncertainty, so the corresponding critical values are larger. As sample size grows the t-distribution converges to the normal, and the two scores become nearly identical. In automated production pipelines, this distinction often does not matter; in formal statistical reporting from small samples it absolutely does.
What are the most common mistakes people make with z-scores?
The first is forgetting that the percentile interpretation requires normality — applying z-table probabilities to a heavily skewed distribution produces nonsense. The second is mixing sample and population parameters: if μ and σ come from the same data you are scoring, your z-scores are slightly distorted (the value you are scoring contributed to its own mean). The third is using z-scores on small samples where t-scores would be more accurate — the resulting probabilities are systematically too small. The fourth is comparing z-scores from two different distributions without checking that both have meaningful means and standard deviations (e.g., highly skewed or multi-modal distributions have z-scores that are arithmetically valid but interpretively misleading). Finally, people sometimes confuse z-scores with raw effect sizes (Cohen's d) — the two have similar form but different denominators.
When should I not use this calculator?
Skip it for non-numeric or ordinal data — z-scores require meaningful arithmetic on the values. Do not use it when σ is unknown and the sample is small (n < 30); switch to a t-score calculator with degrees-of-freedom adjustment. Avoid using the standard normal percentile interpretation on skewed or multi-modal distributions; the same z does not correspond to the same probability. It is the wrong tool for time-series anomaly detection without further work — autocorrelation distorts the apparent z-score. Do not use it as a stand-alone "is this value an outlier?" check without sample-size and distribution context; in a small sample of 10 normal points, a z = 2 is not unusual; in a sample of 10,000, it appears about 500 times by chance. For robust outlier detection use modified z-scores (based on median and MAD) rather than the classic z.