While semantic memory deficits are a common landmark of Alzheimer's disease, the nature of these impairments remains to be clarified. Implicit tasks which assess semantic priming effects are often used to understand semantic deficits in Alzheimer's disease, but they have led to unclear conclusions because of methodological problems such as intervention of attentional mechanisms. To explore the effects of semantic priming in Alzheimer's disease and their relationship with semantic memory deficits, we used two tasks, one implicit and the other explicit. The implicit task was a lexical decision task to assess semantic priming, and in which pairs of words had coordinate (tiger-lion) or attribute relationships (zebra-stripe). The explicit task was a semantic knowledge task composed of namings and questions involving superordinate categories and attribute knowledge of concepts. The two tasks systematically assessed the integrity of the same concepts. This protocol was given to 53 Alzheimer's disease patients with mild to moderate dementia and to 20 controls. The Alzheimer's disease group as a whole obtained significantly greater priming effects (hyperpriming) than controls in the coordinate condition, and equivalent priming in the attribute condition. In the coordinate condition, a subgroup of 26 patients, with attribute knowledge deficits, had larger priming effects than both a subgroup without semantic deficits and the control group. These results show that in Alzheimer's disease the semantic priming effects vary according to the degree of attribute loss, and the presence of hyperpriming would reflect semantic memory deficits. This study unravels the fine-grained structure of semantic memory disturbances in Alzheimer's disease with mild to moderate dementia, affecting initially the attributes of concepts within a hierarchical network in which superordinate concepts remain preserved.