This paper proposes a simple maximum likelihood regression estimator that outperforms Least Squares in terms of efficiency and mean square error for a large number of skewed and/or heavy tailed error distributions