Abstract Objective This longitudinal study used FDG-PET imaging to predict and monitor cognitive decline from normal aging. Methods Seventy-seven 50–80-year-old normal (NL) elderly received longitudinal clinical examinations over 6–14 years (561 person-years, mean per person 7.2 years). All subjects had a baseline FDG-PET scan and 55 subjects received follow-up PET exams. Glucose metabolic rates (MRglc) in the hippocampus and cortical regions were examined as predictors and correlates of clinical decline. Results Eleven NL subjects developed dementia, including six with Alzheimer's disease (AD), and 19 declined to mild cognitive impairment (MCI), on average 8 years after the baseline exam. The baseline hippocampal MRglc predicted decline from NL to AD (81% accuracy), including two post-mortem confirmed cases, from NL to other dementias (77% accuracy), and from NL to MCI (71% accuracy). Greater rates of hippocampal and cortical MRglc reductions were found in the declining as compared to the non-declining NL. Conclusions Hippocampal MRglc reductions using FDG-PET during normal aging predict cognitive decline years in advance of the clinical diagnosis. Future studies are needed to increase preclinical specificity in differentiating dementing disorders.