This paper explores the DSM-IV latent structure of cannabis users (especially its invariance towards gender and age) and assesses the psychometric properties of the Cannabis Abuse Screening Test (CAST) by confrontation with the theoretical diagnoses [dependence and cannabis use disorders (CUD)] and the latent class structure of the DSM-IV. The random sample comprised 550 French cannabis smokers aged 15-62 years interviewed by telephone. DSM-IV diagnoses were assessed with the Munich Composite International Diagnostic Interview. Internal structures of both instruments were assessed using factor analysis and latent class analysis. Optimal CAST cutoffs were determined by sensitivity, specificity and area under the receiver operating curve (AUC). CAST and DSM-IV were unidimensional (Cronbach's alpha = 0.742 and 0.752, respectively), although a two-factor solution showed a better fit for the CAST. CAST cutoffs for screening CUD and dependence were three (AUC = 0.851) and five (AUC = 0.868), respectively. DSM-IV latent class structure varied only marginally in age and gender. Three classes of cannabis smokers were determined, ordered along a continuum of symptoms: non-symptomatic (61.1%), moderate (32.9%) and severe (6.0%). CAST cutoff scores for screening moderate/severe and severe were, respectively, three (AUC = 0.869) and eight (AUC = 0.952). Results are compared to those obtained in previous CAST studies and discussed in line with the DSM-5. Copyright © 2015 John Wiley & Sons, Ltd. [résumé auteur]