Z-score Calculator

A z-score, also called a standard score, is the number of standard deviations a value lies away from the mean of its dataset. It is written z = (x − μ) / σ and is read simply: a z of 1.5 is one and a half standard deviations above the mean.

The transformation was introduced by the statistician Abraham Wald, and its power comes from removing units. The same score describes a student scoring 85 where the class average is 70, and a patient whose blood pressure is 30 points above their own baseline.

Step one is subtraction. Taking x − μ centres the value on zero: the mean itself scores 0, and values above and below the mean fall on either side of it.

Step two is division by the standard deviation. This rescales the gap so that it is measured in typical spreads rather than raw units. Without this step the score would change if you switched from kilograms to pounds.

Step three is optional but powerful: the standard normal table, or Φ(z), converts the score into a percentile rank. Because the normal curve has a fixed shape, a given z always corresponds to the same share of the distribution, whatever the data were measured in.

The two-sided tail probability, 2[1 − Φ(|z|)], answers a related question: how often would a value at least this far from the mean appear by chance in normal data? That is the p-value a z-test would report.

The sign tells you the direction and the magnitude tells you the distance. z = −1.8 is a value 1.8 standard deviations below the mean, not a negative or bad result.

The percentile rank is the share of the distribution at or below the value. A score of 84.13% means the value is higher than about 84% of normally distributed data.

The band label is a quick sanity check: within one standard deviation covers about 68% of normal data, one to two covers a further 27%, two to three about 4.3%, and beyond three roughly 0.3%.

The two-sided tail probability is the same number a z-test would call a p-value. Below 0.05 means the value would be unusual in normal data; it does not by itself prove anything is wrong.

• The standard deviation. Because it sits in the denominator, an understated spread inflates every score, and a value measured with a noisy instrument gets a larger z than it deserves.
• The shape of the data. The percentile reading assumes normality, so skewed data with the same z does not correspond to the same rank in the observed sample.
• Sample size is not part of the formula. A z-score describes a single value against a spread, so unlike a test statistic it does not grow more convincing as more data are collected.
• Outliers distort both inputs. One extreme point raises the standard deviation, which compresses every other score toward zero and hides the outlier instead of exposing it.

• Comparing results across different scales, such as scores in different exams, or measurements in different units.
• Spotting outliers in a dataset, especially with the sample or population standard deviation taken directly from the list.
• Quality control, where a measurement is expressed in multiples of the process standard deviation from its target.
• Describing a result in a report: “2.3 standard deviations above the mean” is far clearer than a raw figure with no context.
• As the building block for further statistics, since the z-test and many standardised scores are built on the same transformation.

• The percentile rank assumes the data are approximately normal. For heavily skewed data, quote the z-score as a distance and use the median and interquartile range for rank information.
• The mean and standard deviation must be meaningful for the comparison. Scoring against a mean from a different population or a different time period produces a number with no interpretation.
• A z-score is not a probability and not a percentage of anything. It is a standardised distance; the percentile reading is the part that refers to a share of the distribution.
• With very small samples the mean and standard deviation are unstable, so the resulting score has a wide margin of uncertainty even though the arithmetic is exact.
• The standard deviation must be greater than zero. If the calculator reports an error, the data have no spread, and scoring against them is not meaningful.