A longitudinal study design is used to find out whether health and health behaviours at age 12-16 predict educational level in early adulthood. The purpose is to study direct (based on health) and indirect (based on health behaviours) health-related selection mechanisms in adolescence. These mechanisms contribute to the allocation of people into various educational positions and thus to the creation of socio-economic health differences in adulthood. Baseline data at age 12-16 from the Adolescent Health and Lifestyle Survey (Finland) in 1981, 1983 and 1985 were linked with data on highest attained education at age 27-33, obtained from the Register of Completed Education in 1998. In the baseline surveys, all 12-, 14- and 16-year-olds born within a specified range of birth dates in July 1964, 1966, 1968, or 1970 were included in the samples (N=11149). The response rate in the mailed surveys varied between 74% and 88% in boys and between 85% and 92% in girls. Associations between baseline variables and attained educational level were assessed by polychotomous logistic regression analysis.Health-compromising behaviours and poor perceived health in adolescence predicted low educational level in adulthood. Several behaviours had independent associations with attained educational level, while associations between health and educational level were mostly accounted for by school achievement and sociodemographic background, which were strong and independent predictors of educational level. The study indicates that in adolescence, indirect selection based on health behaviours, rather than direct selection by perceived health, contributes to the production of socio-economic health differences.