A long-term follow-up was made of a cohort of 307 heroin addicts admitted into a high-dose, Australian methadone maintenance programme in the early 1970s. Using data from clinic records, official death records and methadone treatment histories, it was found that subjects were nearly three times as likely to die outside of methadone maintenance as in it (95% CI RR 1.45 to 5.61). Data were further analyzed using Cox regression to investigate the association of maximum daily methadone dose and a change in clinic policy with retention in maintenance treatment. It is estimated that subjects given a maximum daily dose of 80 mg were nearly twice as likely to be discharged during the first three years of maintenance as those given 120 mg (95% CI RR 1.3 to 2.2). The estimated median time in maintenance for subjects given a maximum dose of 120 mg was 1150 days while for 80 mg it was 660 days. It is further estimated that the change in clinic policy from abstinence to indefinite maintenance reduced to one-third subjects' risk of leaving after three years' of treatment (95% CI RR 0.19 to 0.54). It is concluded that, in order to minimize heroin addicts' risk of death, they should be offered indefinite, high-dose methadone maintenance.