One of the principal barriers to overcoming addiction is the propensity to relapse, even after months or years of abstinence. Relapse can be precipitated by cues and contexts associated with drug use; thus, decreasing the conditioned properties of these cues and contexts may assist in preventing relapse. The predictive power of drug cues and contexts can be reduced by repeatedly presenting them in the absence of the drug reinforcer, a process known as extinction. The potential of extinction to limit relapse has generated considerable interest and research over the past few decades. While pre-clinical animal models suggest extinction learning assists relapse prevention, treatment efficacy is often lacking when extinction learning principles are translated into clinical trials. Conklin and Tiffany (Addiction, 2002) suggest the lack of efficacy in clinical practice may be due to limited translation of procedures demonstrated through animal research and propose several methodological improvements to enhance extinction learning for drug addiction. This review will examine recent advances in the behavioural and pharmacological manipulation of extinction learning, based on research from pre-clinical models. In addition, the translation of pre-clinical findings-both those suggested by Conklin and Tiffany () and novel demonstrations from the past 13 years-into clinical trials and the efficacy of these methods in reducing craving and relapse, where available, will be discussed. Finally, we highlight areas where promising pre-clinical models have not yet been integrated into current clinical practice but, if applied, could improve upon existing behavioural and pharmacological methods.