A unique dataset collected from the personnel records of a large compa
ny is used to study the relationship between on-the-job training and w
orker productivity. The analysis shows how information contained in a
company database is useful for eliminating heterogeneity bias in the e
stimation of training's impact on wages and job performance. Even when
selection bias in assignment to training programs is eliminated, trai
ning is found to have a positive and significant effect on both wage g
rowth and the change in job performance scores, thereby confirming the
robustness of the relationship between training and productivity.