AIMS: To compare the effectiveness and cost-effectiveness of unobserved versus observed dosing of patients seeking treatment of heroin dependence. DESIGN: Randomized controlled trial and cost-effectiveness analysis. Setting Specialist out-patient drug treatment centres in Australia. PARTICIPANTS: Heroin users seeking maintenance treatment. INTERVENTION: Participants were allocated randomly to observed or unobserved dosing for 3 months. All subjects received buprenorphine-naloxone and weekly clinical reviews. MEASUREMENTS: Primary end-points were retention in treatment and heroin use at 3 months. Costs of treatment were measured (in Australian dollars, AU$) and cost-effectiveness compared. Secondary outcomes included quality of life, psychological symptoms and use of non-opioid drugs. FINDINGS: A total of 119 subjects were randomized and analysed. At 3 months, 33/58 (57%) randomized to unobserved treatment, and 37/61 (61%) observed were retained (log-rank chi2 = 0.04, df = 1, P = 0.84). On an intention-to-treat analysis, reductions in days of heroin use in the preceding month, from baseline to 3 months, did not differ significantly; 18.5 days (95% CI: 21.8-15.3) and 22.0 days (95% CI: 24.3-19.7), respectively (Mann-Whitney U = 807.5, P = 0.13). The mean cost for the unobserved group was AU$1,663 (95% CI 1308-2017) per treatment episode, significantly less than the mean cost for the observed group at AU$2,138 (95% CI 1713-2562). CONCLUSIONS: Retention and heroin use was not significantly different between observed and unobserved dosing groups. Attendance for observed dosing was not associated with worse retention. Treatment with close clinical monitoring, but no observation of dosing, was significantly cheaper and therefore significantly more cost-effective.