Aims. To examine the direction of the longitudinal association between vulnerability for psychosis and cannabis use throughout adolescence. Design. Cross-lagged path analysis was used to identify the temporal order of vulnerability for psychosis and cannabis use, while controlling for gender, family psychopathology, alcohol use and tobacco use. Setting. A large prospective population study of Dutch adolescents [the TRacking Adolescents' Individual Lives Survey (TRAILS) study]. Participants. A total of 2120 adolescents with assessments at (mean) age 13.6, age 16.3 and age 19.1. Measurements. Vulnerability for psychosis at the three assessment points was represented by latent factors derived from scores on three scales of the Youth Self-Report and the Adult Self-Report, i.e. thought problems, social problems and attention problems. Participants self-reported on cannabis use during the past year at all three waves. Findings. Significant associations (r = 0.12-0.23) were observed between psychosis vulnerability and cannabis use at all assessments. Also, cannabis use at age 16 predicted psychosis vulnerability at age 19 (Z = 2.6, P < 0.05). Furthermore, psychosis vulnerability at ages 13 (Z = 2.0, P < 0.05) and 16 (Z = 3.0, P < 0.05) predicted cannabis use at, respectively, ages 16 and 19. Conclusions. Cannabis use predicts psychosis vulnerability in adolescents and vice versa, which suggests that there is a bidirectional causal association between the two. © 2012 The Authors, Addiction © 2012 Society for the Study of Addiction.