There is a vast empirical literature of the effects of training on wages that are taken as an indirect measure of productivity. This paper is part of a smaller literature on the effects of training on direct measures of industrial productivity. We analyse a panel of British industries between 1983 and 1996. Training information (and other individual productivity indicators such as education and experience) is derived from a question that has been asked consistently over time in the Labour Force Survey. This is combined with complementary industry-level data sources on value added, wages, labour and capital. We use a variety of panel data techniques (including system GMM) to argue that training significantly boosts productivity. The existing literature has underestimated the full effects of training for two reasons. First, it has tended to treat training as exogenous whereas in reality firms may choose to re-allocate workers to training when demand (and therefore productivity) is low. Secondly, our estimates of the effects of training on wages are about half the size of the effects on industrial productivity. It is misleading to ignore the pay-off firms take in higher profits from training. The effects are economically large. For example, raising the proportion of workers trained in an industry by 5 percentage points (say from the average of 10% to 15%) is associated with a 4 per cent increase in value added per worker and a 1.6 per cent increase in wages.