This method of statistical sampling, a more refined version of simple random sampling, is used when it becomes necessary for subcategories of the population to be represented fairly.

The population is first divided into homogenous groups, or strata, where the whole is heterogenous—for example, school years in the student population, where the object is to sample grades achieved in various exams. There are two approaches from this point: either the entire population is divided into strata, each of which is then given equal weighting (leading to disproportionate samples if the groups differ greatly in size), or, more properly, the number of samples drawn from each group is proportionate to the stratum's fraction of the whole.

To illustrate this last:

           |  Year 7  |  Year 8  |  Year 9  ||  Total  |
| Students |   166    |   181    |   153    ||   500   |

If the total required samples number 100, the number of samples taken from stratum 1—students in Year 7—will be 1/5 (since the sample size, 100, is one fifth of the total number of students) of 166. The process is repeated for the other strata, rounding fractional results up or down while still maintaining the total sample size. Once the size of each sample has been determined, subjects are isolated using simple random sampling.