INTRODUCTION: Hippocampus volumetry is a useful surrogate marker for the diagnosis of Alzheimer's disease (AD). Our purpose was to compare visual assessment of medial temporal lobe atrophy made by radiologists with automatic hippocampal volume and to compare their performances for the classification of AD, mild cognitive impairment (MCI) and cognitively normal (CN). METHODS: We studied 30 CN, 30 MCI and 30 AD subjects. Six radiologists with two levels of expertise performed two readings of medial temporal lobe atrophy. Medial temporal lobe atrophy was evaluated on coronal three-dimensional T1-weighted images using Scheltens scale and compared with hippocampal volume obtained using a fully automatic segmentation method (Spearman's rank coefficient). RESULTS: Visual assessment of medial temporal lobe atrophy was correlated with hippocampal volume (p < 0.01). Classification performances between MCI converter and CN was better using volumetry than visual assessment of non-expert readers whereas classification of AD and CN did not differ between visual assessment and volumetry except for the first reading of one non-expert (p = 0.03). CONCLUSIONS: Visual assessment of medial temporal lobe atrophy by radiologists was well correlated with hippocampal volume. Radiological assessment is as good as computer-based volumetry for the classification of AD, MCI non-converter and CN and less good for the classification of MCI converter versus CN. Use of Scheltens scale for assessing hippocampal atrophy in AD seems thus justified in clinical routine.