Using a unique pan-European dataset, we rely on two alternative measures of over-education and control stepwise for four groups of covariates in order to interpret the over-education wage penalty in light of theoretical models. Firstly, it appears that a significant fraction (i.e. between 1/5 and 1/3) of PhD holders in Europe are genuinely over-educated. Secondly, these genuinely over-educated PhD holders are found to face a substantial wage penalty (ranging from 15 to almost 30%) with respect to their well-matched counterparts. Finally, unconditional quantile regressions highlight that the over-education wage penalty among PhD holders increases greatly along the wage distribution.