What Is a Z Score?
A z score is a number that says how many standard deviations a value sits from the mean of its distribution. A z score of +1.5 means one and a half standard deviations above average; −0.75 means three quarters of a standard deviation below it. That is the whole definition — everything else is using it.
Why such a simple number runs half of statistics
Raw values carry their units and their context with them: 85 marks, 182 centimetres, 28.5 degrees. You cannot compare them, rank them across subjects, or say which is more unusual without knowing each distribution's centre and spread. The z score strips all of that away. It converts any value, from any approximately normal distribution, into a position on one shared ruler — and because the ruler is shared, a single table (the z score table) converts any z score into a probability, a proportion or a percentile. A z of 1.5 is the 93.32nd percentile whether it came from an exam, a height chart or a thermometer. That portability is why z scores appear in education, psychology, medicine, quality control and finance alike: everyone needed the same trick, and this is it.
The interpretation scale
- z = 0: exactly average. The 50th percentile. Perfectly ordinary.
- z between −1 and +1: the everyday middle — 68.26% of all values live here. Unremarkable in the best sense.
- z = ±1: one standard deviation out — the 84.13th percentile going up, the 15.87th going down.
- z = ±1.645: the 95th / 5th percentile fences. Screening cutoffs live here.
- z = ±1.96: the 97.5th / 2.5th fences — the boundary of the middle 95%, and the most used significance cutoff in science.
- z = ±2: outside the middle 95.44% (which spans −2 to +2). "Unusual" starts here in most textbooks.
- z = ±2.576: the 99.5th / 0.5th fences — the middle 99% boundary.
- z beyond ±3: rarer than 3 in 1,000 under the model. In practice, treat these as flags: either a genuinely exceptional case or a data error asking to be checked. In quality control, a process throwing values beyond ±3σ is considered out of control almost by definition.
Notice what the scale never says: "good" or "bad". A z of +2.5 on a test is excellent; the same z on a blood-pressure chart is a referral. The number reports standing, and the subject matter supplies the judgement. Reports that call a high z score "better" without that context are telling you less than they appear to.
Three examples from real life
Exams — standing inside a class
A mark of 85 where the mean is 70 and the standard deviation is 10 gives z = 1.50. The mark itself ("85") only means something inside that class; the z score travels — it says the student outperformed 93.32% of peers, a statement that survives translation to any other test, school or year.
Height — standing inside a population
At 182 cm against a population mean of 170 cm and standard deviation of 7 cm, z = 1.71: taller than 95.64% of the reference population. Growth charts for children run on exactly this machinery, with the reference mean and spread varying by age and sex — the z score is how "tracking along the 75th percentile line" is computed under the hood.
IQ — a scale built from z scores
IQ scales are standardised to mean 100 and standard deviation 15 by construction, so every IQ point is a z score in disguise: IQ 130 is z = +2.00 (97.72nd percentile), and IQ 85 is z = −1.00 (15.87th percentile). The famous "130 = gifted cutoff" is simply "top 2.28%" wearing friendlier clothes — one reason psychologists also keep the unit normal table, the body-and-tail printing of the same curve, closest to hand.
What a z score is not
- Not a probability. z = 1.96 is a position. The probability attached to it (0.9750 to the left) comes from the table, not from the score itself.
- Not a percentage. A z of 0.90 is not "90%". It is nine tenths of a standard deviation above the mean — the 81.59th percentile, as it happens.
- Not a verdict on the distribution's shape. You can compute z scores for skewed data, but the percentile readings from the normal table then stop applying. The score describes distance honestly either way; the normal interpretation is the part with conditions.
- Not the same as a standard score everywhere. In education, "standard score" is an umbrella that includes z scores, T scores (mean 50, SD 10) and stanines — all linear re-dressings of z. When a report says standard score, check which costume it is wearing: T = 50 + 10z converts in one line.
Where z scores sit in the bigger map
Z scores are the front door to inferential statistics. A test statistic in a z-test is a z score — one computed from a sample mean using the standard error — and it is judged against the same curve, with critical values at 1.645, 1.96 and 2.576 as the fences. Confidence intervals are built by walking a critical number of standard errors out from an estimate. The empirical rule (68–95–99.7) is just the z scale read at whole numbers. Learn the z score properly and a third of an introductory statistics course becomes variations on a theme you already know; that is why this site spends a whole page on its formula and another on calculating it step by step.
Sources and further reading
The definition and properties used on this page follow the standard treatment in the NIST/SEMATECH e-Handbook of Statistical Methods, the US national metrology institute's reference for applied statistics.
Frequently asked questions
What does a z score of 0 mean?
The value equals the mean exactly. It is the 50th percentile — as many values below as above. Far from being a failure state, z = 0 is the most representative score a distribution has.
Can a z score be negative?
Yes — any value below the mean gets a negative z score, which is roughly half of all values. The sign gives direction (below), the size gives distance. Probabilities attached to negative z scores come from the negative table.
What is a good z score?
There is no universal "good" — it depends on the direction the subject rewards. On a test, higher is better and z above +1 is strong; for a golf score or a blood marker, lower may be better; for a manufacturing measurement, the best z score is 0, dead on target. Anyone quoting a "good z" owes you the context.
Is a higher z score always better?
No. Higher means further above the mean, nothing more. Whether above-average is desirable is a question about the thing being measured, not about the statistic.
What z score is considered an outlier?
A common working rule flags |z| above 3 (some fields use 2.5, or 3.5 for large datasets). Under a normal model only about 0.27% of genuine values lie beyond ±3, so extremes deserve a look — but "flag for review" is the correct response, not automatic deletion.
What is the z score range?
Theoretically unlimited in both directions — the formula never refuses an input. In practice, values from real normal-ish data almost always fall between −4 and +4, and printed tables stop at ±3.99 because the remaining tail beyond that is smaller than 0.003%.