Background Temperament characteristics have been suggested to be associated with mental health outcomes, especially depression, but the direction of the association is unknown. In this study, we tested whether temperament characteristics, as defined by the Buss-Plomin adulthood emotionality-activity-sociability (EAS) temperament model, predict depressive symptoms or whether depressive symptoms predict changes in temperament characteristics. Methods Participants comprised a population-based sample of 719 men and 1020 women from the Young Finns study aged 20-35 years at baseline in 1997 and who responded to repeated surveys of temperament and depressive symptoms in four study phases from 1997 to 2012. The associations were tested using linear regression models, repeated cross-lagged structural equation models, parallel latent growth curve models and two-dimensional continuous-time state space model (Exact Discrete Model). Results Both low sociability (β=-0.12, p<0.001) and high negative emotionality (β=0.34, p<0.001) predicted subsequent increased depressive symptoms, whereas earlier depressive symptoms predicted increased negative emotionality (β=0.50, p<0.001), but not low sociability. Limitations The depressive symptoms scale applied may not be used for measuring clinically recognized depression. Conclusions Our findings suggest that the direction of the association is from low sociability to depressive symptoms rather than the reverse, but the association between negative emotionality and depressive symptoms seems to be reciprocal.