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.
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%.