Objective The prevalence of both low testosterone levels and depression increases with age. Currently, there is no consensus regarding the existence of an association. Our study analyses the cross-sectional association of testosterone levels with depressive symptoms and its prospective association with the development of incident depressive symptoms. Design Longitudinal population-based study; based on the data of the Longitudinal Aging Study Amsterdam (LASA) including 608 men aged ≥65 years (median age 75·6 years). Measurements Linear and logistic regression between total and free testosterone levels and depressive symptoms as measured by the Center of Epidemiologic Studies Depression (CES-D) scale, taking into account medical and lifestyle factors. Cox Proportional Hazards model was used to assess incident depressive symptoms. Results Unadjusted linear regression between square-root transformed CES-D scores and free testosterone levels showed a significant inverse association as a continuous variable (β = -0·10, P < 0·05), lowest quartile compared to highest (β = 0·12, P < 0·05) and with a threshold value of 170 pmol/l (β = 0·13, P < 0·05). The results remained significant for the group below threshold after adjustment for all confounders (β = 0·09, P < 0·05). Cox Proportional Hazards Model showed a decreased risk for incident depressive symptoms for men with higher free testosterone levels [HR = 0·997 CI (0·995-1·000)]. Men with the threshold value below 220 pmol/l were at increased risk of incident depressive symptoms [HR = 1·989 CI (1·173-3·374)]. Conclusions Free testosterone levels below 170 pmol/l are associated with depressive symptoms, while free testosterone levels below 220 pmol/l (lowest quintile of our population) predict the onset of depressive symptoms. © 2010 Blackwell Publishing Ltd.