We empirically evaluate wage returns to company training using representative individual-levelcross-sectional data for the years 2000, 2003, and 2007. A comparison group approach allowscomparing wages of participants with non-participants, who were willing to participate intraining, yet were restrained out of random reasons. For training participants, we identify a 7.5%wage premium compared to non-participants, which vanishes once the comparison group isrestricted to employees enrolled for training who finally declined participation. The resultsindicate that typically measured returns to training programs may in fact be returns tounobserved characteristics such as innate ability, personality or cognitive and non-cognitiveskills.