INTRODUCTION AND AIMS: Experiencing previous non-fatal overdoses have been identified as a predictor of subsequent non-fatal overdoses; however, few studies have investigated the association between previous non-fatal overdose experiences and overdose mortality. We examined overdose mortality among injecting drug users who had previously been attended by an ambulance for a non-fatal heroin overdose. DESIGN AND METHODS: Using a retrospective cohort design, we linked data on non-fatal heroin overdose cases obtained from ambulance attendance records in Melbourne, Australia over a 5-year period (2000-2005) with a national death register. RESULTS: 4884 people who were attended by ambulance for a non-fatal heroin overdose were identified. One hundred and sixty-four overdose deaths occurred among this cohort, with an average overdose mortality rate of 1.20 per 100 person-years (95% CI, 1.03-1.40). Mortality rate decreased 10-fold after 2000 coinciding with widely reported declines in heroin availability. Being male, of older age (>35 years) and having been attended multiple times for previous non-fatal overdoses were associated with increased mortality risk. DISCUSSION AND CONCLUSIONS: As the first to show a direct association between non-fatal overdose and subsequent overdose mortality, this study has important implications for the prevention of overdose mortality. This study also shows the profound effect of macro-level heroin market dynamics on overdose mortality.