Aims: To examine the course and the predictors of the persistence of cannabis dependence. Methods: Through cannabis outlets and chain referral, a prospective enriched community cohort of 207 young adults (aged 18-30) with DSM-IV cannabis dependence at baseline (T0) was formed and followed-up after 1.5 (T1) and 3 (T2) years. The presence of cannabis dependence, cannabis-related problems, functional impairment and treatment was assessed using the Composite International Diagnostic Interview (CIDI 3.0) and the Sheehan Disability Scale (SDS). Predictors of persistence were lifetime cannabis abuse and dependence symptoms, cannabis use characteristics, distant vulnerability factors (e.g. childhood adversity, family history of psychological/substance use problems, impulsivity, mental disorders), and proximal stress factors (recent life events, social support). Results: Four groups were distinguished: persistent dependent (DDD: 28.0%), stable non-persistent (DNN: 40.6%), late non-persistent (DDN: 17.9%) and recurrent dependent (DND: 13.5%). At T2, persisters (DDD) reported significantly more (heavy) cannabis use and cannabis problems than non-persisters (DNN/DDN/DND). Treatment seeking for cannabis-related problems was rare, even among persisters (15.5%). The number (OR = 1.23 (1.03-1.48)) and type (‘role impairment' OR = 2.85 (1.11-7.31), ‘use despite problems' OR = 2.34 (1.15-4.76)) of lifetime cannabis abuse/dependence symptoms were the only independent predictors of persistence with a total explained variance of 8.8%. Conclusions: Persistence of cannabis dependence in the community is low, difficult to predict, and associated with a negative outcome. The substantial proportion of stable non-persisters suggests that screening and monitoring or low-threshold brief interventions may suffice for many non-treatment-seeking cannabis-dependent people. However, those with many lifetime abuse/dependence symptoms may benefit from more intensive interventions.