Longitudinal surveys are valuable tools for investigating health and social outcomes across the life course. In such studies, selective mortality leads to changes in the social composition of the sample, but little is known about how selective survey participation affects the sample composition, in addition to the selective mortality. In the present paper, we followed a Swedish cohort sample over six waves 1968-2011. For each wave we recalculated the distribution of baseline characteristics in the sample among i) the sample still alive and ii) the sample still alive and with complete follow-up. The results show that the majority of the compositional changes in the cohort were modest and driven mainly by mortality. However, for some characteristics, class in particular, the selection was considerable and in addition, was substantially compounded by survey non-participation. We suggest that sample selections should be taken into account when interpreting the results of longitudinal studies, in particular when researching social inequalities.