BACKGROUND: Studies on persistence of benzodiazepine agonist (BZDA) withdrawal in older outpatients are few, and few studies on long-term persistence over years have yet been published. To describe the persistence of temazepam, zolpidem, and zopiclone (BZDA) withdrawal among older outpatients at 3 years from the beginning of withdrawal, as well as any changes in use of other medications. METHODS: 92 outpatients (≥55 years) with primary insomnia, long-term BZDA use as hypnotics (mean duration of BZDA use 9.9 ± 6.2 years), and willingness to withdraw from BZDAs each received either melatonin or a placebo nightly for one month. During this period, BZDAs were meant to be gradually withdrawn. Sleep hygiene counselling and psychosocial support were provided. Three years later, use of BZDAs and other medications was determined by interview and confirmed from medical records. RESULTS: Of the original 92 outpatients, 83 (90%) participated in the 3-year survey (mean follow-up 3.3 ± 0.2 years). The number of BZDA-free participants decreased from 34 (37%) at 6 months to 26 (28%; intention-to-treat) at 3 years, that of irregular BZDA users decreased from 44 (48%) at 6 months to 27 (29%) at 3 years, while that of regular users increased from 11 (12%) at 6 months to 30 (33%) at 3 years (P = 0.001). Those who were regular BZDA users at 3 years had at baseline (before withdrawal) higher BMI (P = 0.001) than did other participants. At 3 years, the total number of medications remained unchanged for non-users (P = 0.432), but increased for the irregular (P = 0.011) and regular users (P = 0.026) compared to baseline. At 3 years, compared to baseline, use of antidepressants, dopamine agonists, melatonin, and NSAIDs/paracetamol was significantly more common in the whole cohort, but their use did not differ between the BZDA-user subgroups. Randomization to melatonin or placebo during BZDA withdrawal was unrelated to BZDA-withdrawal result. CONCLUSIONS: At 3 years after withdrawal, the number of BZDA-free participants had decreased, but still one-third of the subjects remained BZDA-free, and one-third had reduced their use. Successful BZDA withdrawal did not lead to any increase in total number of medications; use of symptomatic medications in the whole cohort, however, did increase.