Neuroimaging studies in healthy subjects have yielded controversial results about the neural substrates of autobiographical memory. Moreover, the neural networks responsible for autobiographical amnesia remain poorly understood. Since autobiographical memory is frequently altered in Alzheimer’s disease (AD), we used this degenerative disorder as a model and applied a correlative approach between resting cerebral glucose utilization (CMRGlc) and temporally graded memory scores to identify the cerebral structures whose synaptic dysfunction subserves the impairment in autobiographical memory. To this end, we studied a group of 17 AD patients with mild to moderate dementia in whom autobiographical memory was assessed using a specially designed task from three broad time periods [the previous 5 years (period A); middle age (period B); and teenage and childhood (period C)], and measures of resting CMRGlc were obtained with PET. The patients performed less well than a control group for all three time periods and showed the expected temporal gradient, with the most remote period being best preserved (Ribot’s gradient). Qualitative analysis showed that remote memories concerned generic (i.e. semantic) rather than specific (i.e. episodic) events. We found a significant positive correlation between autobiographical scores and the metabolism of the right hippocampus (extending to the lingual gyrus), restricted to period A. In addition, period A scores were significantly correlated with the right middle and inferior frontal gyri and the right middle temporal gyrus. Period B scores correlated chiefly with the prefrontal cortex bilaterally (bilateral superior, bilateral middle and right inferior gyri). Metabolic correlations with period C scores were restricted to the left middle frontal gyrus. These findings show striking differences in metabolic correlations with the autobiographical time period, in agreement with prevalent theories of normal functioning of human memory. Thus, in accordance with theories of long-term memory consolidation, we find the expected implication of the hippocampal region in the recall of recent memories, and a disengagement of this structure when the retention interval is beyond 5 years. Moreover, according to the hemispheric encoding/retrieval asymmetry model based on activation studies in healthy subjects, the fact that recent memories preferentially involved the right prefrontal cortex whereas remote memories involved the left prefrontal cortex supports the notion of semanticization of memories with time interval, such that preserved remote memories in AD have a predominantly semantic character.