In this paper we discuss the interpretation of the specific robust M-estimator that has become popular in applied econometrics, henceforth termed BWM-estimator, and give the conditions required for it to be consistent for the parameters of the conditional mean. In particular, we emphasize that in the presence of skewed heteroskedastic errors this M-estimator will be inconsistent for these parameters and note that its efficiency can be severely affected by heteroskedasticity. Although we focus on the BWM-estimator, our results extend to other robust regression estimators as it is illustrated both in the simulations and in the empirical application we present.