We evaluate a method to describe changing substance use patterns in northern Australia’s remote Aboriginal communities (Arnhem Land, Northern Territory). Substance use was assessed in random samples in two communities A (n=194) and B (n=176). Five Aboriginal health workers made assessments independently of each other in community A. A different group of three health workers made independent assessments in community B. Sub-samples were opportunistically recruited for interview (community A, n=77; community B, n=55). In community C, 101 people were interviewed and were also assessed by four local health workers working together. Proportional agreements (kappa- k statistic) among health workers for a history of substance use and current use, varied from k =0.207 for petrol sniffing (P=0.006) up to k =0.749 for cannabis use (P<0.001), all better than would be expected by chance. In communities A and B, agreement between health workers’ consensus and self-reported substance use was weaker (0.103< k <0.482) probably because of under-reporting in interviews. In community C, where interviews were conducted in a confidential clinic setting, agreement between health workers’ consensus and self-report varied from k=0.273 for petrol sniffing (P<0.001) up to 0.819 for tobacco use (P<0.001). Aboriginal health worker consensus classification clarified equivocal self-report data.