Abstract
The classes of R- and M-estimates contain practical robust alternatives to least squares estimation in linear models. These estimates form the basis for a robust analysis of variance. This inference procedure is described and its versatility demonstrated.