Medial temporal atrophy is a well-established marker for Alzheimer's disease (AD). However, due to normal variation in the size of medial temporal structures and variability in how radiologists interpret images, the use of clinical reads in establishing the presence of pathological atrophy is imprecise. A limitation of studies of magnetic resonance imaging (MRI) measures in AD is diagnostic uncertainty as it can be unknown if pre- or early-symptomatic subjects go on to develop AD and most subjects do not undergo autopsy verification of the diagnosis. In persons with or at-risk for AD due to fully-penetrant autosomal dominant mutations in the PSEN1 and APP genes, the diagnosis or future development of AD can be predicted with essentially 100% accuracy. We used this predictability to assess the ability of radiologists to detect hippocampal atrophy (HA) in persons destined to develop AD. Coronal T1-weighted MRI scans of 39 persons demented from (n = 4) or at-risk for inheriting (n = 35) PSEN1 or APP mutations were independently assessed by two radiologists and the presence or absence of HA determined. Of the 39 subjects, 26 were FAD mutation carriers. Fifteen of 28 asymptomatic at-risk persons were FAD mutation carriers and four of these were rated as having atrophy for a sensitivity of 27% and a specificity of 85%. Among seven mildly affected yet non-demented subjects, atrophy was detected in three and in the four demented subjects HA was identified in two. Our results suggest that radiologists' ability to detect HA in persons in whom the diagnosis of incipient AD is certain is sub-optimal and quantitative MRI techniques or other biological markers of the disease are needed.