This study examines the ability of amnesic patients to recover newly formed associations implicitly after a single study trial. Fifteen amnesic patients with various etiologies studied pairs by forming a sentence containing both words. At test, all participants saw 40 intact pairs, 40 rearranged pairs, and 40 new words. All pairs appeared side by side both at study and at test. For the implicit lexical-decision task, 40 nonwords were intermixed with the other pairs, and participants indicated whether both items were words. For the explicit speeded-recognition test, participants were asked to indicate whether both words had appeared at study. Despite being severely impaired on the explicit test, amnesic patients performed like healthy controls on the implicit test, with faster and more accurate responses to intact pairs than to recombined pairs. Contrary to existing theories, the results suggest that amnesic patients can form and retain new associations.