Amnesia is a key component of Alzheimer's disease (AD) and the most important feature of its clinical diagnosis but its specificity has recently been challenged. This study investigated the ability of amnesia to predict AD in a clinicopathological dementia series. Ninety-one patients to which free and cued verbal memory assessment was administered during early cognitive decline, were followed until autopsy. Patients' histological diagnoses were classified as pure AD, mixed AD, and non-AD pathologies. Data-driven automated classification procedures explored the correspondence between memory performance and pathological diagnoses. Classifications revealed 3 clusters of performance reflecting different levels of amnesia. Little correspondence between these clusters and the presence of AD pathology was retrieved. A third of patients with pure/mixed AD pathology were non-amnesic at presentation and ≈45% of patients without AD pathology were amnesic. Data-driven prediction of AD pathology based on memory also had a poor accuracy. Free and cued memory assessments are fair tools to diagnose an amnesic syndrome but lack accuracy to predict AD pathology.