A substantial literature argues that the reported job dissatisfaction of union members is spurious. It reflects either the sorting of dissatisfied workers into union membership or the sorting of union recognition into worse jobs. We contribute by presenting the first panel data estimates that hold constant first worker fixed effects and then worker in job fixed effects. The estimates demonstrate that covered union members typically report greater dissatisfaction even when accounting for sorting in both dimensions. At the same time, cross-sectional evidence of less satisfied covered non-members and uncovered members vanishes when accounting for sorting.