The Relapse Replication and Extension Project (RREP) was a multisite study to replicate and extend Marlatt's taxonomy of relapse precipitants. In addition to replicating Marlatt's original taxonomic system, three independent research teams utilized prospective designs to identify additional predictors of relapse and developed and evaluated two alternative systems for assessing high risk relapse situations. This overview describes the replication methodology, summarizes seven RREP studies completed by the three research groups, and discusses five cross-cutting conclusions emerging from the studies. These conclusions are: (1) reliability of Marlatt's taxonomic system was variable both within and across the three research sites; (2) Marlatt's taxonomic system showed little predictive validity in analyses that used pretreatment relapse data to predict post-treatment relapse, but there are important unresolved issues; (3) an alternative taxonomy provided little more predictive validity than the original taxonomy even though it measured more dimensions of relapse situations and provided greater analytic flexibility; (4) the Reasons for Drinking Questionnaire appeared to be a successful psychometric transformation of Marlatt's taxonomy, one which did demonstrate predictive validity; and (5) Marlatt's taxonomy was based on a time-intensive model of relapse prediction whereas RREP prospective analyses represented time-extensive models of relapse prediction. Coping responses are noted to be effective predictors of relapse under both models.