Background: Few studies describe cannabis use in indigenous populations, and no longitudinal studies are available in Australia. We conducted 3-year follow-up interviews and assessments in Aboriginal communities in Arnhem Land (Northern Territory, NT). Methods: A randomly selected sample (n = 161; 80 males, 81 females aged 13-36 years) was assessed in October 2001 and then reassessed in September 2004. An opportunistically recruited sample (n = 104; 53 males, 51 females aged 13-36 years) was also interviewed in 2001 and followed-up in 2004. Cannabis and other substance use were determined by combining proxy assessments by local Aboriginal health workers, medical records and data from interviews. Changes in cannabis use and symptoms of misuse were assessed using McNemar's test for paired proportions and the Wilcoxon signed rank test. Logistic regression assessed associations between clinical presentations and cannabis use at both time-points. Results: Those who used cannabis at both baseline and follow-up were at greater risk than those who never used it to have suffered: auditory hallucinations; suicidal ideation; and imprisonment. In the randomly selected cohort there were fewer cannabis users at follow-up than at baseline (P = 0.003). The reduction was evident in females generally (P = 0.008) and older males (aged = 16 at baseline) (P = 0.007). In those interviewed at both baseline and follow-up we measured no statistically significant reduction in frequency and levels of use, although fewer cannabis users reported symptoms of misuse such as: fragmented thought processes; memory disruption; difficulties controlling use; and auditory and visual hallucinations. Conclusions: Modest reductions in cannabis use and its consequences in this population were demonstrated. These may be the result of enhanced supply control and broader socio-political changes. � 2006 Society for the Study of Addiction.