To examine the relative contribution of fluency and recollection to the word completion performance of amnesics, we administered a task in which patients were told specifically not to utilize previously presented words during stem completion (an Exclusion condition). This condition was contrasted with a standard word completion task in which patients were encouraged simply to complete the stem with the first word that came to mind (an Inclusion condition). Since the exclusion condition necessitated controlled respecification of the initial presentation, it was hypothesized that amnesics would be less able than controls to exclude study list items. Consistent with this hypothesis, the results indicated that the amnesics' performance, unlike that of the alcoholic controls, did not significantly differ as a function of task condition. To examine whether amnesics' conscious recollection could be enhanced, Experiment 2 presented the study list five times. The amnesics now were able to exclude a significant number of items from the study list; however, they still did so considerably less frequently than alcoholic controls. For the alcoholic controls, increasing the number of study trials had little additional effect on their exclusion performance, but it significantly enhanced their inclusion performance. Taken together, these findings suggest that for control subjects, word completion performance is likely mediated by a combination of fluency and recollection, while for amnesic patients, performance is almost exclusively based on the fluency with which an item comes to mind.