Abstract
This paper utilizes a rich data set on workers and their employers in the US and Japan to test several predictions of human capital theory. The data set incorporates both prospective and retrospective measures of turnover, includes multiple measures of training, and provides a basis for calculating plant-specific returns to tenure. Contrary to human capital theory, there is no evidence that establishments with high levels of training have either high returns to tenure or low levels of turnover. Surprisingly, establishments with high returns to tenure do not have low levels of turnover.