Z-Score Calculator

Background.

The Quanta z-score calculator turns any raw observation into a standard score in a single click: enter the value x, the population mean μ, and the population standard deviation σ, and the tool returns z = (x − μ) / σ along with the matching percentile under the standard normal curve and a two-sided p-value. Standardisation is the most important idea in elementary statistics — it strips the units out of a measurement and replaces them with a single dimensionless number that says, in the universal language of standard deviations, exactly how unusual the observation is. A test score of 130, a stock daily return of 4%, a baby weighing 4.1 kg at birth, and a six-sigma manufacturing defect rate are all completely incomparable on their native scales, but their z-scores live on the same number line and can be ranked, averaged, plotted, and fed into the same hypothesis tests without any further bookkeeping.

The standard normal distribution N(0, 1) is the destination of that transformation. It is bell-shaped, symmetric about zero, and characterised by a probability density φ(z) = (1/√(2π)) exp(−z²/2) and a cumulative distribution Φ(z) that has no closed-form elementary expression but is tabulated in every statistics textbook and computed numerically in every spreadsheet. The empirical rule (also called the 68-95-99.7 rule) reads directly off Φ: about 68.27% of a normal population sits within ±1σ of the mean, about 95.45% within ±2σ, and about 99.73% within ±3σ. The slightly different number every student memorises — z = 1.96 for a 95% confidence interval — is what you get when you ask for the exact two-sided 95%, not 95.45%: Φ(1.96) − Φ(−1.96) = 0.95000, so 1.96 is the precise critical value that leaves 2.5% in each tail. The 1.645 you also remember is the one-sided 95% critical value (Φ(1.645) = 0.95).

These two numbers, 1.96 and 1.645, anchor a huge fraction of all applied statistics: every textbook confidence interval of the form x̄ ± z·σ/√n, every introductory hypothesis test against a normal null, every quality-control chart with control limits at ±3σ, and every Cohen's d effect size all live or die by them.

Z-scores are not just classroom abstractions; they drive consequential decisions in the wild. The Stanford-Binet and Wechsler IQ tests are by construction normed to N(100, 15), so an IQ of 130 is exactly z = 2 and the Mensa cutoff at the 98th percentile is a hair under z = 2.05. Pediatric growth charts published by the CDC and the WHO report a child's height-for-age, weight-for-age, and BMI-for-age as z-scores so that paediatricians can compare a four-month-old in Nairobi to a four-month-old in Boston without first converting kilograms to pounds. Finance uses z-scores to compute Sharpe ratios — the excess return of a portfolio divided by its standard deviation is a z-score in disguise — and credit scoring tools such as Altman's Z-Score combine several financial ratios into a single distress-prediction score that bankruptcy judges have admitted into evidence. Six Sigma manufacturing, the discipline that made General Electric in the 1990s, is named for the goal of holding defect rates to fewer than 3.4 per million opportunities, which corresponds to a process capability of roughly z = 6 on the relevant tolerance band.

Behind every one of those applications is the same three-line calculation: subtract the mean, divide by the standard deviation, look up the tail probability. This calculator handles all three steps with the Abramowitz-Stegun §26.2.17 rational approximation to the normal CDF, accurate to better than 7.5 × 10⁻⁸ across the whole real line — the same approximation Excel, R, and most pocket calculators use under the hood — so the percentile and p-value you see here will match any statistics package or textbook table to four decimal places and usually to six. The explainer below covers when to use a z-score versus a t-statistic, how to read one-tailed and two-tailed p-values without falling into the classic 'doubled p' trap, why z = 1.96 and not z = 2 anchors the 95% confidence interval, and how the empirical rule lets you ballpark percentile estimates in your head without ever opening a table.

What is z-score calculator?

A z-score, also called a standard score or z-value, is the number of standard deviations a single observation lies above or below the mean of its distribution. Formally, for an observation x drawn from a population with known mean μ and known standard deviation σ, the z-score is z = (x − μ) / σ. The transformation has two simultaneous effects: it shifts the distribution so its mean is zero, and it scales it so its standard deviation is one. After standardisation the observation lives in the standard normal distribution N(0, 1) regardless of the original units, provided the parent distribution was normal to begin with. A positive z indicates the observation is above the mean, a negative z that it is below the mean, and z = 0 that it sits exactly at the mean. The magnitude of z is a unitless measure of how unusual the observation is: roughly 68% of a normal population has |z| ≤ 1, roughly 95% has |z| ≤ 1.96, and roughly 99.7% has |z| ≤ 3. Z-scores are most useful when (a) the parent distribution is approximately normal and (b) σ is known, not estimated from a small sample. When σ is unknown and must be estimated from the data via the sample standard deviation s, the z-score is replaced by a t-statistic t = (x̄ − μ) / (s/√n), which follows a Student's t-distribution with n−1 degrees of freedom rather than a standard normal — the t-distribution has fatter tails to compensate for the extra uncertainty introduced by estimating σ. For samples larger than about 30, the t-distribution converges on the standard normal and the two procedures give nearly identical answers, which is why introductory texts often blur the distinction. The percentile associated with a z-score is the cumulative probability Φ(z) = P(Z ≤ z) where Z is a standard normal random variable. Φ has no closed-form elementary expression — it is defined as an integral of the normal PDF from −∞ to z — so values are obtained either from tabulated z-tables, from numerical approximations such as Abramowitz & Stegun §26.2.17 (used here), or from rational/Chebyshev approximations built into statistical software. The two-sided tail probability, 2·(1 − Φ(|z|)), is the p-value for testing the null hypothesis that the observation was drawn from N(μ, σ²) against the alternative that it was drawn from some distribution with a different mean.

How to use this calculator.

1. 01Enter the raw observation x in the first field. This is the value you want to standardise — a test score, a measurement, a return, a height. The number can be positive, negative, or zero, and any units are fine as long as the mean and standard deviation below are in those same units.
2. 02Enter the population mean μ. For standardised tests like the SAT or the Wechsler IQ test the mean is set by the test designers (500 and 100 respectively); for a real-world measurement you usually plug in either a known population mean or a sample mean x̄ that you have computed elsewhere.
3. 03Enter the population standard deviation σ. This must be strictly greater than zero — a distribution with σ = 0 collapses to a single point and z is undefined. If you only have a sample standard deviation s from a small dataset (n < 30) you should be using a t-statistic and the Quanta t-test calculator instead; for n ≥ 30 the s ≈ σ approximation is fine in practice.
4. 04Read the z-score directly. A z of 2 means the observation is two standard deviations above the mean; a z of −1.5 means one and a half below. Use the 68-95-99.7 rule for a quick reality check: |z| > 2 is rare (about one in twenty observations) and |z| > 3 is genuinely unusual (about three per thousand).
5. 05Read the percentile to translate the z-score into a population ranking. A percentile of 97.72% means the observation sits at or above 97.72% of the population, equivalently in the top 2.28%. Percentiles are the form non-statisticians find easiest to digest — say 'top 2%' rather than 'two sigma above the mean' in any audience that does not eat distributions for breakfast.
6. 06Use the two-sided p-value when you are running a hypothesis test against H₀: μ = μ₀ with a two-sided alternative. Compare it to your significance level α (typically 0.05 or 0.01); reject H₀ if p < α. If your alternative is one-sided you want half of this number — divide it by two, since the symmetric normal puts equal probability in each tail.

The formula.

The z-score itself is one of the cleanest formulas in statistics:

z = (x − μ) / σ

where x is the observation, μ is the population mean, and σ > 0 is the population standard deviation. The numerator (x − μ) is the raw deviation in the original units; the division by σ rescales that deviation so that one full unit of z always corresponds to one full standard deviation of the parent distribution. The transformation is linear, so it preserves the shape of the distribution — if the parent is normal, the result is the standard normal N(0, 1); if the parent is right-skewed, the standardised distribution is still right-skewed, just centred at zero with unit variance.

The percentile and the two-sided p-value require the cumulative distribution function Φ(z) of the standard normal:

Φ(z) = (1/√(2π)) · ∫_{−∞}^{z} exp(−t²/2) dt

This integral has no closed-form elementary expression, so it is computed numerically. This calculator uses the rational approximation from Abramowitz & Stegun §26.2.17, the same approximation that ships inside Microsoft Excel, the GNU Scientific Library, and most statistical packages:

For z ≥ 0: k = 1 / (1 + 0.2316419·z) P(k) = a₁·k + a₂·k² + a₃·k³ + a₄·k⁴ + a₅·k⁵ Φ(z) ≈ 1 − φ(z)·P(k)

For z < 0: Φ(z) = 1 − Φ(−z)

where φ(z) = (1/√(2π))·exp(−z²/2) is the standard normal PDF and the coefficients are a₁ = 0.319381530, a₂ = −0.356563782, a₃ = 1.781477937, a₄ = −1.821255978, a₅ = 1.330274429. Abramowitz & Stegun report a worst-case absolute error of |ε(z)| ≤ 7.5 × 10⁻⁸ across the entire real line, which is overkill for any percentile reported to four decimal places.

The percentile is then

percentile = Φ(z) × 100

and the two-sided tail probability is

p = 2·(1 − Φ(|z|))

The absolute value in |z| folds the negative-z half of the curve onto the positive half by symmetry; the factor of 2 then accounts for the equally extreme observation in the opposite tail. For a one-sided test, take p/2 directly.

The widely-memorised critical values fall straight out of this machinery. Solving Φ(z) = 0.975 gives z = 1.95996 — rounded to 1.96, the two-sided 95% critical value that anchors every introductory confidence interval of the form x̄ ± 1.96·σ/√n. Solving Φ(z) = 0.95 gives z = 1.64485 — the one-sided 95% critical value. The 68-95-99.7 rule is just Φ(1) − Φ(−1) ≈ 0.6827, Φ(2) − Φ(−2) ≈ 0.9545, and Φ(3) − Φ(−3) ≈ 0.9973 rounded to one significant figure.

A worked example.

Frequently asked questions.

References& sources.

1. [1]Abramowitz, M., & Stegun, I. A. (Eds.). (1972). Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables (Dover, 9th printing). §26.2.17 — rational approximation to the normal CDF Φ(x) with maximum absolute error 7.5 × 10⁻⁸.
2. [2]NIST/SEMATECH e-Handbook of Statistical Methods, §1.3.6.6.1 — Normal Distribution. National Institute of Standards and Technology.
3. [3]Casella, G., & Berger, R. L. (2002). Statistical Inference (2nd ed., Duxbury). Chapter 3 — Common families of distributions, including the standard normal and the derivation of Φ.
4. [4]Pearson, K. (1894). Contributions to the Mathematical Theory of Evolution. Philosophical Transactions of the Royal Society of London A, 185, 71–110 — origin of the term 'standard deviation' and early treatment of the normal distribution.
5. [5]Zwillinger, D. (Ed.). (2018). CRC Standard Mathematical Tables and Formulas (33rd ed., CRC Press). Standard reference tables for the normal CDF Φ(z) and inverse Φ⁻¹(p).
6. [6]WHO Multicentre Growth Reference Study Group. (2006). WHO Child Growth Standards: Length/height-for-age, weight-for-age, weight-for-length, weight-for-height and body mass index-for-age — methods and development. World Health Organization, Geneva.
7. [7]Altman, E. I. (1968). Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy. The Journal of Finance, 23(4), 589–609 — original Altman Z-Score paper.
8. [8]Sharpe, W. F. (1966). Mutual Fund Performance. The Journal of Business, 39(1), 119–138 — introduces the Sharpe ratio as a z-score-style reward-to-variability measure.