This paper investigates the e®ects of training on labor productiv- ity using a unique nationally representative panel of Italian ¯rms for the period 2002 to 2005. We ¯nd that training has a positive and signi¯cant e®ect on productivity. Using a variety of panel estimation techniques, we show that failing to account for unobserved heterogene- ity leads to overestimate the impact of training on productivity, while failing to account for endogeneity leads to substantially underestimate it. Training also has a positive and signi¯cant impact on wages, but this e®ect is about half the size of the e®ect on productivity. Within occupational groups, the e®ect of training on productivity is large and signi¯cant for blue-collars, but small and not signi¯cant for white collars.