Sample sizes are small ($n < 15$) and the data are skewed - Heavy outliers are present (these inflate the parametric test's standard error) - The data are ordinal (means may not be meaningful) - The distributions have very different shapes across groups