The performance of three kinds of amnesic patients and control subjects was assessed using four methods for testing memory: free recall, recognition, cued recall, and word completion. Whereas amnesic patients were impaired on free recall, recognition, and cued recall, they were normal on word completion. Moreover, performance on the word-completion test declined at a normal rate reaching chance after about 120 min. The word-completion test resembled the cued-recall test in that the initial letters of previously presented words were given as cues. It differed from cued recall only in the instructions, which directed subjects away from the memory aspects of the test and asked them to complete each three-letter cue with the first word that came to mind. The present results offer an explanation of conflicting findings that have been obtained with amnesic patients on tests of the cued-recall type. The results are considered in terms of a process (activation or procedural learning), which is spared in amnesia and not dependent on the integrity of the damaged brain regions.