A test has mean 70 and SD 10. Ana scores 85, which is 15 points above the mean, and 15 points is one and a half standard deviations: z=1085−70=1.5. The z-score rewrites her score as a distance from the mean, counted in SDs. A score of 60 would be z=−1, one SD below the mean, and a score of exactly 70 would be z=0.
Using the SD as the unit is what makes different tests comparable. Being 15 points above average is remarkable on a test where most scores sit within 5 points of the mean (z=3), and ordinary on one where scores routinely swing by 30 (z=0.5). The z-score builds that context into the number. It also lets every normal distribution share one table: whatever the original mean and SD, z=1.5 falls at about the 93rd percentile.
The raw deviation x−μ looks similar but is not a z-score. It is still in the original units (15 points) and cannot be compared across tests. A percentile is different again: it is the share of values below a point, and for normal data you get it from z, but z=1.5 and the 93rd percentile are different numbers describing the same score. Always standardize with the mean and SD of the value's own group, never with numbers borrowed from another data set.
Think of it like this. Converting to z-scores is like converting prices into one currency before comparing them: dividing by the SD turns 'points on this test' into a shared unit, standard deviations. The conversion is only as good as the data's shape allows; for strongly skewed data, the same z can sit at quite different percentiles.
Core idea. A z-score restates a value as the number of standard deviations it sits above or below the average.