The Oaxaca-Blinder decomposition is widely used in empirical studies to explain changes in wage or income distributions. It separates an observed difference between two groups into an explainable part due to individuals’ characteristics and an unexplainable part. It is possible to draw a parallel between the Oaxaca-Blinder decomposition and the treatment effect literature (Rubin 1974, Imbens and Rubin 2015). We exploit this parallel to apply recent techniques measuring the treatment effect. Thus, we extend the method to be robust to functional form misspecification of regression models, using recent machine learning methods (Chernozhukov et al. 2018, Athey et al. 2019). An illustration investigates the decomposition of the gender pay gap.