Objective: This study attempted to compare two models for selective (people at elevated risk) and indicated (those with subsyndromal depressive symptoms) prevention and to determine the optimal strategy for prevention of late-life depression. Method: Onset was assessed at 3 years with the Geriatric Mental State AGECAT in a randomly selected cohort of 1,940 non-depressed and nondemented older people in Amsterdam. Risk factors that can easily be identified in primary care were used. Results: The association of risk factors with depression incidence was expressed in absolute and relative risk estimates, number needed to treat, and population-attributable fractions. Prevention models were identified with classification and regression tree analyses. In the indicated prevention model, subsyndromal symptoms of depression were associated with a risk of almost 40% of developing depression and a number needed to treat of 5.8, accounting for 24.6% of new cases. Adding more risk factors raised the absolute risk to 49.3%, with a lower number needed to treat but also lower attributable fraction values. In the selective prevention model, spousal death showed the highest risk, becoming even higher if the subjects also had a chronic illness. Overall, the attributable fraction values in the indicated model were higher, identifying more people at risk. Conclusions: Consideration of the costs and benefits of both models in the context of the availability of evidence-based preventative interventions indicated that prevention aimed at elderly people with depressive symptoms is preferred. The focus on treatment should be readdressed; a new approach is needed, with a stronger emphasis on prevention.