OBJECTIVE To estimate the relative rates of dementia in initially nondemented subjects with and without memory impairment defined by baseline free recall from the Free and Cued Selective Reminding (FCSR) test. BACKGROUND Our approach to identifying persons at high risk for future dementia is to show the presence of memory impairment not caused by other cognitive deficits by using a memory test that controls attention and cognitive processing. When the conditions of testing are not adequately controlled, prediction is reduced because age-associated memory deficits due to other cognitive deficits are confused with dementia-associated memory deficits. METHODS Longitudinal evaluation of 264 initially nondemented, elderly community volunteers from the Einstein Aging Study with clinical and psychometric examinations every 12 to 18 months for up to 10 years. METHODS Dementia was defined by an algorithmic definition that required a Blessed Information Memory and Concentration score >8 and clinical evidence of functional decline. RESULTS Thirty-two incident cases of dementia developed during follow-up. Survival analyses indicated that subjects with impaired free recall at baseline had dementia develop (relative risk = 75.2, 95% CI = 9.9 to 567) over 5 years of follow-up at dramatically higher rates than subjects with intact free recall after adjusting for age, gender, and education. CONCLUSIONS Poor performance on free recall from FCSR predicts future dementia. These findings support the existence of a preclinical phase of dementia characterized by memory impairment, which is present for at least 5 years before diagnosis.