BACKGROUND AND AIMS: Heroin overdose is a serious consequence of heroin use and one of the leading causes of premature death and illness in Australia. Despite considerable research effort little is known about the effects of transient changes in heroin user behaviour and the links to overdose. This research is the first to use a suitable methodology to allow such ephemeral changes and their effects on non-fatal heroin overdose to be examined. METHODS: A case-crossover design was used in which non-fatal heroin overdose survivors' recall of risk behaviours in the 12 hours prior to overdose (hazard period) was compared to their recall of risk behaviours in the 12 hours prior to a selected non-overdose heroin injection (control period). RESULTS: A total of 155 participants were able to provide valid details of hazard and control periods. A dose-response relationship was observed between the self-reported amount of heroin used and likelihood of overdose (e.g. > AUD50, OR 12.97, 95% CI 2.54-66.31). The use of benzodiazepines (OR 28, 95% CI 3.81-205.79) or alcohol (OR 2.88, 95% CI 1.29-6.43), during the hazard period was related to overdose risk, but the effect of alcohol was attenuated by the effect of benzodiazepines. Shifting from private to public locations between control and hazard periods was also related to increased risk of overdose (OR 3.63, 95% CI 1.66-7.93). CONCLUSIONS: We demonstrate the value of a new methodology to explore heroin overdose, as well as discussing its limitations and ways to overcome them in future. In terms of our findings, overdose prevention messages need to highlight the impact of these transient changes in behaviour and to emphasize the risks of using higher doses of heroin as well as continuing to emphasize the risks of combining heroin with other central nervous system (CNS) depressants. Safer environments for heroin use, such as injecting rooms, may also reduce the chances of overdose.