Accelerated forgetting has been proposed as the first sign in preclinical and early Alzheimer's disease (AD). The authors investigated learning and retention in participants who later developed AD with free and cued selective reminding (FCSR; H. Buschke, 1984; E. Grober & H. Buschke, 1987), a test that maximizes learning by inducing deep semantic processing and by controlling study and test conditions. AD patients in the preclinical stage recalled significantly fewer words than did matched control participants, indicating an impairment of learning; nonetheless, patients' retention was identical to that of control participants. A retention deficit was documented 3 years later for AD patients but not for control participants, whose retention was still perfect. Thus, a retention deficit is not present in preclinical AD when hallmark learning deficits can be documented. Detection of preclinical and very early AD may be best accomplished by using robust learning tests that control cognitive processing.