This paper studies the relationships between on-the-job training, wages and job performance by using the personnel records of a large manufacturing firm. Utilizing a company database avoids the biases that generally result when individuals are unable to accurately recall the amount of training they received and/or when definitions of training vary across diverse firms. The main findings presented in this paper are: (1) Controlling for information on days spent in formal training programs reduces the returns to tenure by 18%; (2) First-difference models of wage growth which eliminate heterogeneity bias in wage levels show that training has a positive and significant effect on wage growth; (3) Fixed-effects models that control for heterogeneity bias in wage growth still find a positive and significant effect of training on wage growth; and (4) Training leads to an improvement in job performance, as measured by performance rating scores.