Aim. While cannabis dependence has been increasingly recognized, there is little research on the measurement issues involved in operationalizing the dependence syndrome for this drug. This paper aimed to investigate the diagnostic utility and appropriate diagnostic cut-offs of three short dependence measures among long-term cannabis users. Setting and participants. Two hundred long-term, regular cannabis users were recruited and interviewed in Sydney, Australia. Measurements. Receiver Operating Characteristic analyses compared the diagnostic performance of the short University of Michigan CIDI, a measure of ICD-10 dependence and the Severity of Dependence Scale against the "gold standard" of moderate or more severe DSM-III-R cannabis dependence, as diagnosed by the Substance Abuse Module of the CIDI. Findings. The measures were of equal utility in diagnosing at least moderate DSM-III-R cannabis dependence. While the optimal diagnostic cut-offs for the short University of Michigan CIDI and the ICD-10 dependence measure remained unchanged from those conventionally applied, a more liberal cut-off was optimal for the Severity of Dependence Scale. The amended prevalence of cannabis dependence was 77% using the short University of Michigan CIDI, 72% by the ICD-10 measure and 62% by the Severity of Dependence Scale. Conclusions. The three instruments were able to diagnose cannabis dependence at levels substantially better than chance. They were generally robust in terms of the optimal diagnostic cut-off in a population of long-term cannabis users. This paper provides guidelines for choosing optimal cut-offs within different contexts.