Taiwan has experienced a rapid expansion in higher education since the 1990s. To gauge changes in earnings returns to higher education caused by this expansion, this paper estimates college effects on earnings using both the conventional Mincer-type regression model and the revised truncated-sample model that adjusts for the selection mechanisms into college. We also apply Xie and Wu's (2005) hierarchical linear model approach to test if the treatment effects of higher education vary as a function of propensity scores strata estimated. Using nationwide data collected in the early 1990s and the early 2000s, we focus on young entrants to the labor market. Our results indicate that average returns to college education remain stable over time. We also find that in both periods, there is a strong negative selection mechanism at work: when workers with a low latent propensity of receiving college education indeed did go to college, they benefit the most from the college attendance.