Many prospective clinical studies have concluded that craving does not reliably predict relapse and that the concept is of little or no clinical utility. Contrary to earlier more simplistic clinical models of addiction, more recent models do not require that craving be present for relapse to occur. New approaches to study human craving may enhance its predictive validity and yield more knowledge of its nature, course, behavioural sequelae and regulatory function in alcohol/drug consumption. These approaches include empirical research that focuses on: (1) the elucidation of the domains of craving (i.e. subjective experience, physiological responses, behavioural sequelae and their inter-relationships); (2) the temporal dynamics of craving (i.e. its course over minutes or days, as well as its natural history over the course of a drinking career); (3) the factors that may mediate/moderate/determine the development and resolution of craving; (4) studies of the predictive validity of craving measures; and (5) the development of valid methods of measuring the different domains of craving. The conclusions are that future craving research should: (1) incorporate more sophisticated general theories of behaviour (conditioning, cognitive social learning, neurobiological, and genetic); (2) apply more sophisticated and standardized measurement methods and experimental paradigms, including studies in which alcohol is made available to human subjects; and (3) effective development of new pharmacological and behavioural therapies for relapse prevention depend on greater understanding of the nature and measurement of craving.