In this commentary on the Relapse Replication and Extension Project, several points are addressed. First, the selection of the pretreatment "baseline relapse" as the key predictor variable for post-treatment relapse in the predictive and construct validity studies is questioned. It is doubtful whether retrospective accounts of drinking episodes after 4 days of abstinence qualify as relapse episodes for the following reasons: clients may not have been committed to abstinence (e.g. weekend binge drinkers); treatment may have included coping skill training for pretreatment high-risk situations (i.e. treatment intervention is viewed as a "nuisance variable"). By assuming that a client's retrospective report of a poorly defined pretreatment "relapse" is predictive of the type of relapse episode experienced during the post-treatment follow-up assessment is to commit an error of "taxidermy" in evaluating the reliability and validity of the taxonomy; i.e. the tendency for researchers to "stuff and permanently mount" distal baseline episodes as static predictor variables, thereby depriving them of their dynamic and fluid role as proximal determinants in the relapse process. The difference between researchers who assess distal trait predictors of relapse and clinicians who focus more on proximal state predictors in conducting relapse prevention therapy is discussed and illustrated by means of a case study. Finally, the metaphor of relapse as "falling off the wagon" is described as an example of a systems approach to relapse prevention.