This paper develops a two-period labor market model with imperfect information and on-the-job training, and uses data from National Longitudinal Survey of Youth 1979 Cohorts (NLSY79) to test its predictions. We find that training does not explain the positive relationship between employer size and wage. In addition, for industries that display size-wage premium, workers in large establishments are more likely to receive on-the-job training but their return to training is smaller. Our theory, substantiated by the new empirical evidence, suggests that it is not large firms, per se, but firms that hire better workers who are paying a wage premium.