This article focuses on the interrelationship between educational mismatch and earnings, taking three new approaches. First, we examine decompositions of the mismatch wage gap, finding that characteristics explain less than half of the mismatch penalty. Second, we use unconditional quantile regression to examine the mismatch penalty across the earnings distribution, showing that the penalty shrinks as the position in the earnings distribution increases. Third, we decompose the differentials using quantile decompositions. Different reasons for mismatch show heterogeneity in our results, with larger penalties for being mismatched due to working conditions, location, family, and no available job.