Tests whether two independent groups differ, working from ranks rather than the values themselves. It assumes no normal distribution, so it holds up with outliers and with ordinal data, standing in where a t-test is hard to justify.