Marlatt's focus on the relapse situation has had a major impact upon research and clinical practice in treating addictions. One component of his work was the development of a taxonomy for classifying precipitants of relapse. This taxonomy has been incorporated into the nomenclature of clinicians and clinical researchers as part of an explanatory framework for understanding relapses. Despite the taxonomy's influence it has never been examined for the reliability of its use across research studies. The present study compared the reliability of independent classifications of 149 relapse episodes by trained raters at three research laboratories. Despite considerable across-laboratory training, reliability was found to be inconsistent for research purposes. It is concluded that comparability of results based on Marlatt's relapse taxonomy across independent studies must be subject to question, and assumptions necessary for the aggregation of a knowledge base are not supported. Recommendations are offered for improving the reliability of the taxonomy and the methods used to collect taxonomy data. More generally, questions regarding the value of the specific relapse categories, as well as the overall taxonomy, are raised.