The use of robust regression has gained popularity among applied econometricians. Unfortunately, most practitioners who have used these estimators seem to be unaware of the fact that their properties can be dramatically affected by both heteroskedasticity and skewness of the errors. In this paper we reconsider the interpretation of a specific robust regression estimator that has become popular in applied econometrics, and conclude that its use in this context cannot be generally recommended. Alternatively, quantile and mode regression could be used when the researcher wants to estimate conditional location functions that are robust to the presence of outliers.