Some heroin addicts persistently fail to benefit from conventional treatments. We aimed to compare the effectiveness of supervised injectable treatment with medicinal heroin (diamorphine or diacetylmorphine) or supervised injectable methadone versus optimised oral methadone for chronic heroin addiction. In this multisite, open-label, randomised controlled trial, we enrolled chronic heroin addicts who were receiving conventional oral treatment (>or=6 months), but continued to inject street heroin regularly (>or=50% of days in preceding 3 months). Randomisation by minimisation was used to assign patients to receive supervised injectable methadone, supervised injectable heroin, or optimised oral methadone. Treatment was provided for 26 weeks in three supervised injecting clinics in England. Primary outcome was 50% or more of negative specimens for street heroin on weekly urinalysis during weeks 14-26. Primary analysis was by intention to treat; data were adjusted for centre, regular crack use at baseline, and treatment with optimised oral methadone at baseline. Percentages were calculated with Rubin's rules and were then used to estimate numbers of patients in the multiple imputed samples. This study is registered, ISRCTN01338071. Of 301 patients screened, 127 were enrolled and randomly allocated to receive injectable methadone (n=42 patients), injectable heroin (n=43), or oral methadone (n=42); all patients were included in the primary analysis. At 26 weeks, 80% (n=101) patients remained in assigned treatment: 81% (n=34) on injectable methadone, 88% (n=38) on injectable heroin, and 69% (n=29) on oral methadone. Patients on injectable heroin were significantly more likely to have achieved the primary outcome (72% [n=31]) than were those on oral methadone (27% [n=11], OR 7.42, 95% CI 2.69-20.46, p<0.0001; adjusted: 66% [n=28] vs 19% [n=8], 8.17, 2.88-23.16, p<0.0001), with number needed to treat of 2.17 (95% CI 1.60-3.97). For injectable methadone (39% [n=16]; adjusted: 30% [n=14]) versus oral methadone, the difference was not significant (OR 1.74, 95% CI 0.66-4.60, p=0.264; adjusted: 1.79, 0.67-4.82, p=0.249). For injectable heroin versus injectable methadone, a significant difference was recorded (4.26, 1.63-11.14, p=0.003; adjusted: 4.57, 1.71-12.19, p=0.002), but the study was not powered for this comparison. Differences were evident within the first 6 weeks of treatment. Treatment with supervised injectable heroin leads to significantly lower use of street heroin than does supervised injectable methadone or optimised oral methadone. UK Government proposals should be rolled out to support the positive response that can be achieved with heroin maintenance treatment for previously unresponsive chronic heroin addicts. Community Fund (Big Lottery) Research section, through Action on Addiction.