Objectives: Craving plays an important role in the development and maintenance of alcohol dependence and in relapse after periods of abstinence. Anti-craving compounds, such as acamprosate, naltrexone or serotonergic compounds, are found to be only moderately effective. These moderate effects might be due to inadequate matching of specific patients to specific treatments. In 1999, Verheul et al. proposed a three-pathway model of craving in alcoholics, which hypothesised that reward drinkers would better respond to naltrexone, relief drinkers to acamprosate and obsessive drinkers to serotonergic compounds. However, these matching hypotheses are not yet validated. This article reviews the literature on predictors and matching variables of the effectiveness of pharmacological interventions in alcohol dependent patients directed at the reduction of craving and the prevention of relapse. Methods: Studies were selected through a literature search in September 2004, focusing on matching or predicting variables for anti-craving compounds in the treatment of alcoholics. Matching or predicting variables were categorised as either phenotypic (e.g., level of baseline anxiety), endophenotypic (e.g., physiological cue reactivity) or genetic (e.g., m opioid receptor polymorphisms). Results: Studies using clinical or phenotypic effect modifiers have produced inconsistent and rather disappointing results. In contrast, the predictive value of genetic effect modifiers are quite promising (e.g., m opioid receptor polymorphisms). No studies that looked at endophenotypic predicting or matching variables were identified. Conclusion: It is concluded that phenotypic matching variables might be too distal, i.e., far removed from the pathogenic process, and that matching research should shift its attention toward more proximal variables (e.g., genetic and endophenotypic variables).