In this study, we explore whether education-occupation mismatch in the form of over-education and under-education affects the risk of long-term sickness absence. We use register data covering the Norwegian population in the period 2003-2013, containing 13,628,079 person-year observations (2,059,989 persons). To account for selection, we controlled for unobserved occupational and individual heterogeneity. This has been lacking in previous studies of mismatch and health outcomes. Our results show that over-education increases, and under-education reduces, the probability of long-term sickness absence. Controlling for selection reduces the associations substantially. The associations also diminish with experience. These results hold across two different measures of mismatch. Furthermore, both time spent in a mismatched status and specifications of mismatch alter the probability of LTSA. We conclude that education-occupation mismatch is associated with long-term sickness absence but underscore the need for adequate controls for selection.