Dunn's test is the not parametric analog of Holm-Sidak multiple t-test. When you use the Kruskal-Wallis test, you know if there is a difference among your groups, but you can't apply the KWtest for each couple because the overall error is >alpha (Bonferroni inequality). By Dunn's test you can highlight where is the difference using multiple, stepdown comparisons. The algorithm requires the Statistics Toolbox.