Background We argue for a translational approach to addiction science, using an important current research question as a case study. Case study What is the evidence in support of the hypothesis that alcohol increases the risk of a heroin/opiate overdose through a pharmacological interaction? Findings The positive epidemiological evidence shows that opiate overdose deaths rarely involve a single drug; that alcohol is the most common other drug involved; that there is a negative association between alcohol and morphine concentration at post mortem; and that post-mortem levels of morphine are often below the levels expected of highly tolerant individuals. The evidence is consistent with the hypothesis that heroin users who drink may require less heroin to overdose than those who do not drink (all other factors being equal) because of a pharmacological interaction. However, the evidence is consistent with, and does not rule out, other causal (and non-causal) pathways. Alcohol could be associated negatively with tolerance, or confounded by other factors. Experimental evidence is required which is unlikely to be obtained through further epidemiological study or through randomized clinical trials. Conclusions We believe that animal models could provide the key evidence to test the hypothesis for a 'pharmacodynamic' or 'pharmacokinetic' interaction, which could be corroborated in clinical challenge studies and epidemiological studies. Such a translational approach demands greater collaboration between addiction scientists from basic to applied science and from neuroscience to social science, and would be able to address other key research questions and hypotheses in addiction.