Z-Score Calculator
Convert any value to its z-score (standard score) given the population mean and a KNOWN population standard deviation. Outputs percentile plus one-tailed and two-tailed tail areas. No significance verdict — that depends on sample size and study design, which this tool does not ask for.
Z-Score Calculator
Why there is no "significant / not significant" verdict here. Statistical significance is not a property of a z-score. It depends on the sample size behind the number, the design of the study, and a hypothesis fixed before the data was seen — none of which this form asks for. A tool that printed a verdict from x, μ and σ alone would be guessing: hold those three inputs fixed and vary n, and the same z can move from unremarkable to overwhelming. So this calculator reports the z-score and its tail areas, and stops there.
What σ has to be. The formula z = (x − μ) / σ assumes σ is a known population standard deviation and x is a single observation. If you paste in a sample standard deviation s — the kind our standard deviation calculator reports — you are substituting an estimate for a parameter, and at small n the result is not a z-score. At n = 3, a value whose true z is 2.00 can land anywhere from roughly 0.9 to 9.2 depending on which sample you happened to draw.
What to use instead. If x is a sample mean rather than one observation, the denominator you want is the standard error σ/√n — start with the standard error calculator, which sets out the SD-versus-SE distinction properly. If σ is unknown and estimated from a small sample, a t-based method is the correct route, not this one.
How to use the Z-Score Calculator
Enter your three values: x, μ, σ
x = the raw value you want to standardise (e.g. a test score of 85, a height of 175 cm, a stock return of 8%). μ (mu) = the population mean (e.g. average test score 70, average height 170 cm, average return 5%). σ (sigma) = the population standard deviation (e.g. SD of test scores = 10, SD of heights = 7 cm, SD of returns = 15%). The calculator instantly computes z = (x − μ) / σ.
σ must be a known population value. That is the assumption baked into the formula, and it is stricter than it sounds. A sample standard deviation s — the number you get from pasting a dataset into a standard deviation calculator — is an estimate of σ, not σ itself, and at small n it is a noisy one. Substituting it here produces a number that looks like a z-score but is not distributed like one: at n = 3, a value whose true z is 2.00 can come out anywhere from roughly 0.9 to 9.2. If σ is unknown, use a t-based method instead of this tool.
Interpret the z-score
The z-score is the number of standard deviations the value is away from the mean. z = 0 means right at the mean. z = +1 means one SD above the mean. z = −2 means two SDs below the mean. The 68-95-99.7 rule: ~68% of normal data has |z| ≤ 1, ~95% has |z| ≤ 2, ~99.7% has |z| ≤ 3. So a z-score of +2.5 is in the top 0.6% of normal data; z = +3 is in the top 0.13%. Note the flip side of the first band: about 32% of normal data — roughly one value in three — sits outside ±1 SD, so |z| just over 1 is ordinary, not unusual.
Read the percentile
Percentile = the percentage of values that fall BELOW your data point in a normal distribution. z = 0 → 50th percentile (median). z = +1 → 84th percentile. z = +2 → 97.7th percentile. z = −1 → 16th percentile. Useful for things like: "this child's height is in the 75th percentile" or "this test score is in the top 5%". The percentile assumes the underlying distribution is approximately normal — if your data is heavily skewed, percentiles from z-scores will be inaccurate.
Read the tail areas — and know what they are not
The two tail-area figures are exact probabilities under the standard normal curve. The one-tailed figure is P(Z > |z|): the share of a normal population lying further from the mean than your value, in one direction. The two-tailed figure doubles it, covering both directions. Two-tailed is the default reading; one direction only makes sense if you fixed that direction before seeing the data.
These are tail areas for a single observation, not the p-value of a study. This tool deliberately prints no "significant / not significant" verdict, because it never asks for the sample size, and significance cannot be decided without it: hold x, μ and σ fixed, change n, and the same inputs can move from unremarkable to overwhelming. If your x is a sample mean, the z you want is (x̄ − μ) / (σ/√n) — get that denominator from the standard error calculator first. If σ was estimated from your data rather than known in advance, a t-based method is the correct route.
Z-scores — the universal currency of statistics
The z-score (also called standard score) is one of the most useful single concepts in statistics. It converts any value from any normally-distributed dataset onto a universal scale: "how many standard deviations from the mean?" This standardisation lets you compare apples and oranges — a student in the 90th percentile on the SAT compared to a 90th-percentile applicant in a national fitness test, despite the underlying scores being completely different scales. The math is simple: z = (x − μ) / σ. The interpretation is universal: positive z = above mean, negative z = below mean, magnitude = how far in units of standard deviation.
From z-score to percentile and back
The standard normal distribution (mean 0, SD 1) has a well-known cumulative distribution function (CDF) Φ(z) that maps any z-score to its corresponding percentile. Key values: Φ(0) = 0.50 (50th percentile, the median). Φ(1) ≈ 0.84 (84th percentile). Φ(2) ≈ 0.977 (97.7th percentile). Φ(3) ≈ 0.9987 (99.87th percentile). The inverse direction is equally useful: to find the value at the 95th percentile, you need z = 1.645 (one-tailed) or z = 1.96 (two-tailed). These two values — 1.96 and 1.645 — are the most-cited z-scores in statistical practice because they correspond to the 5% significance threshold for two-tailed and one-tailed tests respectively. Memorising them pays off across every quantitative discipline.
Z = 1.96 (two-tailed) and Z = 1.645 (one-tailed) are the most-cited z-scores in statistics. They correspond to the 5% significance threshold — the universal "p < 0.05" cutoff.
P-values: useful but often misunderstood
The p-value is the probability of observing data at least this extreme IF the null hypothesis is true. A common misinterpretation is that p < 0.05 means "5% chance of being wrong" — this is incorrect. P-values say nothing about the probability of the null hypothesis being true; they say only how surprising the observed data would be assuming H₀. Sound usage: a small p-value is evidence AGAINST the null hypothesis but doesn't quantify how much. The American Statistical Association issued a 2016 statement explicitly warning against the mechanical use of p < 0.05 as a decision rule. Effect sizes (standardised mean differences like Cohen's d) and confidence intervals are now considered more informative for practical interpretation. But the z-score → p-value path remains the foundation of frequentist statistical testing.
The ASEAN data-science + statistics angle
Statistical literacy across ASEAN has surged with the rise of data-science education. Z-scores show up routinely in: A/B testing for tech companies (Grab, Shopee, Lazada, GoTo, Tokopedia all run thousands of A/B tests monthly; z-tests determine which variants ship); medical research (clinical trials at NUS, NUHS, SingHealth, KKH publish in Lancet / NEJM using standard frequentist inference); fintech credit scoring (Sea Group, GXS Bank, OCBC, DBS use z-scores in risk models); educational psychometrics (SAT-equivalent national exams across ASEAN report z-scores or percentiles); quality control in manufacturing (Singapore's electronics + biopharma SPC dashboards). For the average APAC professional moving into a data role, mastering z-scores + standard normal distribution is the first chapter of any inferential statistics course. This calculator handles the math; understanding when and how to apply z-tests vs t-tests vs other tests is the lifelong skill that builds on top of it.
10 Things to Know About Z-Scores
Z-score formula: z = (x − μ) / σ. Number of standard deviations the value x is from the mean μ.
The standard normal distribution has mean = 0 and SD = 1. Z-scores convert any normally-distributed data onto this universal scale.
Z = 1.96 (two-tailed) and Z = 1.645 (one-tailed) correspond to the 5% significance threshold (p < 0.05).
The 68-95-99.7 rule: ~68% of normal data has |z| ≤ 1; ~95% has |z| ≤ 2; ~99.7% has |z| ≤ 3.
The standard normal CDF Φ(z) maps z-scores to percentiles. Φ(0) = 0.50, Φ(1) ≈ 0.841, Φ(2) ≈ 0.977, Φ(3) ≈ 0.9987.
Z-scores assume the underlying distribution is roughly normal (bell-shaped). For heavily skewed data, percentiles from z-scores are inaccurate.
The p-value is the probability of data at least this extreme under the null hypothesis. NOT the probability the null hypothesis is true.
Use a two-tailed test by default; one-tailed only when you have a directional hypothesis stated BEFORE looking at data.
"6 Sigma" puts the spec limit 6 SD from the process mean. Its familiar 3.4-per-million figure is the 4.5σ tail, not the 6σ one — the method allows a 1.5σ long-run drift. An un-shifted two-sided 6σ tail is about 2 per billion.
The 2016 ASA statement on p-values explicitly warns against mechanical use of p < 0.05 as a decision rule. Always report effect sizes alongside p-values.
Frequently Asked Questions
-
A z-score (or "standard score") is the number of standard deviations a value is away from the mean of its distribution. Formula: z = (x − μ) / σ. Positive z = above the mean; negative z = below the mean. Magnitude tells you how far. The standardisation lets you compare values from different distributions on a common scale — a z-score of +2 means "2 SD above the mean" whether you're measuring test scores, height, weight, stock returns, or anything else. It's the lingua franca of inferential statistics.
-
Depends entirely on context. For test scores: higher z-score = better. For pollutant levels: lower z-score = better. For health markers (cholesterol, blood pressure): z-scores near zero (closer to typical) are usually best. Magnitude interpretation: |z| < 1 covers about 68% of normal data, so it is entirely typical. |z| between 1 and 2 is ordinary too — about 32% of normal data, roughly one value in three, falls outside ±1 SD. |z| between 2 and 3 is genuinely uncommon (about 4.6% of the distribution, one in 22). |z| > 3 is rare (about 0.27%, one in 370) and worth investigating as an outlier or a strong signal. The familiar thresholds |z| > 1.96 (5%, two-tailed) and |z| > 2.58 (1%) are cut-offs for a properly specified z-test — one where the z was computed against the correct standard error for the quantity being tested. They are not a verdict you can read off a single standardised observation, which is why this tool reports tail areas and leaves the inference to you.
-
Use TWO-tailed by default. Use one-tailed only when you have a strong directional hypothesis stated BEFORE looking at the data — "this drug LOWERS blood pressure" (not "changes" it). Two-tailed tests whether your data is significantly different from the mean in EITHER direction (both higher and lower count as evidence). One-tailed tests only one direction. The one-tailed p-value is exactly half the two-tailed p-value, so one-tailed reaches significance "more easily" — this is why analysts sometimes incorrectly use one-tailed post-hoc to fish for significance. Pre-registering your hypothesis direction is the only legitimate way to use one-tailed.
-
p < 0.05 means: if the null hypothesis were true, the probability of observing data at least this extreme would be less than 5%. It does NOT mean: there's a 95% chance the alternative hypothesis is true (a common misinterpretation). It does NOT mean: the effect is large or practically important (small effects can be highly significant with large samples). It does NOT mean: the result will replicate (replicability requires effect size + sample size context). The 2016 American Statistical Association statement explicitly warned against mechanical interpretation. Always report effect sizes (Cohen's d, % difference, etc.) alongside p-values.
-
Because for the standard normal distribution, Φ(1.96) = 0.975 (97.5th percentile). The remaining 2.5% in each tail sums to 5% total — exactly the 5% significance level for a two-tailed test. So if your |z| > 1.96, you're in the most extreme 5% of values expected under the null hypothesis. The corresponding one-tailed value is z = 1.645 (Φ(1.645) = 0.95). These numbers are worth memorising — they show up across every quantitative discipline. For more stringent tests: z = 2.58 → p < 0.01 (two-tailed); z = 3.29 → p < 0.001.
-
Use a z-test when you KNOW the population standard deviation σ. Use a t-test when you ESTIMATE the SD from your sample (the more common case). The t-distribution has wider tails than the normal distribution to account for the extra uncertainty from estimating SD; the difference is larger for small samples. As n grows, the t-distribution converges to the normal distribution — for n > 30, they're virtually identical. For small samples (n < 30), always use t-test. For large samples (n > 30) with unknown σ, either works (t-test is technically correct but z-test is a good approximation). When σ is genuinely known (rare in practice), use a z-test — and note that this tool assumes exactly that case. Either way, if you are testing a MEAN rather than a single observation, the denominator is the standard error, not the raw SD: the standard error calculator works out SE = σ/√n and explains why the two are different quantities.
-
Z-score percentile interpretations require normal-ish data. For heavily skewed data (income, response times, ecological measurements), z-scores still calculate but percentiles will be wrong. Options: (1) Transform the data first (log, square root, Box-Cox) to make it more normal. (2) Use non-parametric tests (Mann-Whitney U, Wilcoxon) that don't assume normality. (3) Compute percentiles directly from the data using empirical ranks — our standard deviation calculator reports the empirical quartiles (Q1, median, Q3) and IQR, which are distribution-free. The Central Limit Theorem helps for large samples — even non-normal data has approximately normal SAMPLE MEANS when n > 30, so z-tests on the mean are usually valid even when individual values aren't normal.
-
Within ±0.00001 for typical z-scores, using the Abramowitz-Stegun rational approximation of the standard normal CDF. That bound has been checked against a high-precision reference across the usable range: the worst observed absolute error was about 7×10⁻⁸, comfortably inside ±0.00001. For extreme z-scores (|z| > 6), the precision degrades slightly but the tail area is essentially zero anyway. For research-grade precision (publication-quality, when |z| > 5 matters), use R / Python / SAS, which carry higher-precision implementations. For 99%+ of practical work, this approximation is more than adequate.
-
No. All calculations run entirely in your browser via JavaScript. There's no server roundtrip — open DevTools → Network and confirm zero outbound requests. Your data stays on your device. Safe for clinical trial analyses, proprietary A/B test results, sensitive research data, or any inferential statistics work that shouldn't leave your machine.
-
Everywhere quantitative. A/B testing: z-tests determine which variant is statistically significant (Shopee, Lazada, Grab, GoTo all run thousands of A/B tests monthly using z or t-tests). Quality control: SPC charts in manufacturing flag readings outside ±3σ. Medical research: clinical trial endpoints use z-tests for normally-distributed outcomes. Finance: Sharpe ratio is essentially a z-score (excess return / SD). Education: SAT, GRE, GMAT scores are normalised to specific mean/SD; percentile reports come from z-scores. Psychology: T-scores (z scaled to mean 50, SD 10) used in personality + IQ testing. The math in this calculator is the foundation of all of these.
Pick up where you left off
Stored only in this browser — never sent to our servers.