This study evaluates the French version of the Cannabis Use Disorders Identification Test (CUDIT) in a clinical sample with a high rate of cannabis misuse. A group of 194 cannabis users, including 134 outpatients (Group 1) seen at a specific cannabis clinic and 60 drug/alcohol addicts hospitalized for withdrawal (Group 2), filled out the CUDIT. The majority (82% of the sample) were cannabis dependent or abusers (DSM-IV). Internal reliability was assessed by Cronbach's alpha, item-total correlation by Spearman's rho and temporal reliability by the test-retest method. Construct validity was evaluated by factor analysis. Convergent validity was evaluated by measuring the correlation with the quantity of cannabis consumed. The cut-off score used to distinguish cannabis users from misusers was set using a Receiver Operating Characteristic curve analysis. Cronbach's alpha was high (0.83). There was no significant difference between the total test and retest scores (p = 0.1) indicating good temporal reliability. Principal component analysis identified 2 principle factors that explained 53.1% of the total variance. Convergent validity was significant and satisfactory. The cut-off value was 10, which corresponded to the optimal sensitivity (94.4%) and positive predictive value (90.4%). However, the item-total correlation and the correlation with abuse and dependence severity were mediocre for item 9. It should be removed because of its poor psychometric properties. The CUDIT is a satisfactory screening instrument for cannabis use disorders in at-risk populations.