While the importance of on-the-job training is recognized by everyone, it is a phenomenon that is very difficult to study. Most training is informal and hard to measure and its effects on productivity are even more difficult to quantify. An elegant theory explaining how the quantity of training is determined and who pays for and benefits from it has been available for more than a third of a century (Becker 1962). However, the absence of data on the key theoretical constructs of the theory--general training, specific training, informal training and productivity growth--means that the only predictions of the theory that have been tested relate to the effects of formal training and tenure (interpreted as a proxy for informal training) on wage growth and turnover. Until recently, definitive tests of the OJT theory were infeasible because the large number of unobservables meant that any given phenomena had many alternative explanations (Garen, 1988). New data sets with improved measures of OJT are at last becoming available and consequently there has been a good deal of progress recently in testing OJT theory. This paper provides a review of this work.