INTRODUCTION: In most longitudinal studies of problem opiate users, drop-outs are frequent, but not taken into account. However, missing data can induce important bias in parameters estimates. OBJECTIVE: The aim of this study was to examine the influence of drop-outs in the statistical analysis of a follow-up of opiate users in maintenance treatment. METHODS: Participants were 519 patients who had sought maintenance treatment between 1994 and 2001. Drug use was studied using the drug composite score of the Addiction Severity Index. A classical data analysis (linear mixed effects model for repeated measurements) was compared with a selection model, which consists, in this case, of a joint modelling of the score and of the drop-out probability in order to reduce bias induced by drop-outs. RESULTS: At 18 months, 38% of the patients were available for evaluation. Drop-outs were associated with low drug use and were informative. Each model showed that the score decreased over time and that it was associated with psychiatric problems. Unlike the classical method, the joint model showed no significant association between the score and age or treatment setting. CONCLUSIONS: These results show the importance of accounting for informative drop-outs in data analysis before drawing conclusions from such studies.