The demographics of aging suggest a great need for the early diagnosis of dementia and the development of preventive strategies. Neuropathology and structural MRI studies have pointed to the medial temporal lobe (MTL) as the brain region earliest affected in Alzheimer's disease (AD). MRI findings provide strong evidence that in mild cognitive impairments (MCI), AD-related volume losses can be reproducibly detected in the hippocampus, the entorhinal cortex (EC) and, to a lesser extent, the parahippocampal gyrus; they also indicate that lateral temporal lobe changes are becoming increasingly useful in predicting the transition to dementia. Fluoro-2-deoxy-D-glucose positron emission tomography (FDG-PET) imaging has revealed glucose metabolic reductions in the parieto-temporal, frontal and posterior cingulate cortices to be the hallmark of AD. Overall, the pattern of cortical metabolic changes has been useful for the prediction of future AD as well as in distinguishing AD from other neurodegenerative diseases. FDG-PET on average achieves 90% sensitivity in identifying AD, although specificity in differentiating AD from other dementias is lower. Moreover, recent MRI-guided FDG-PET studies have shown that MTL hypometabolism is the most specific and sensitive measure for the identification of MCI, while the utility of cortical deficits is controversial. This review highlights cross-sectional, prediction and longitudinal FDG-PET studies and attempts to put into perspective the value of FDG-PET in diagnosing AD-like changes, particularly at an early stage, and in providing diagnostic specificity. The examination of MTL structures, which has so far been exclusive to MRI protocols, is then examined as a possible strategy to improve diagnostic specificity. All told, there is considerable promise that early and specific diagnosis is feasible through a combination of imaging modalities.