Post-Lasso and Post-Double-Lasso are becoming the most popular meth-ods for estimating average treatment e˙ects from linear regression models with many covariates. However, these methods can su˙er from substantial omitted variable bias in finite sample. We propose a new method called Post-Double-Autometrics, which is based on Autometrics, and show that this new method outperforms Post-Double-Lasso when explanatory variables are weakly correlated with the endogenous variable but correlated with the treatment variable.