This paper presents for the first time panel evidence on the productivity effects of training intensity and different training forms in Germany. It hereby takes account of selectivity of training activities, unobserved heterogeneity of establishments as well as omitted variable bias. Using the waves 1997 – 2000 of the IAB establishment panel, it is found that when the share of trained employees in 1997 is higher, productivity is significantly higher in the period 1997 - 1999. Formal internal and external courses have the highest positive impact on productivity, self-induced learning and quality circles have a smaller positive impact, while training on the job, seminars and talks and job rotation do not affect productivity. The decision to train is selective. Firms with an inefficient production structure deliberately use training in order to boost productivity.