We used principal component analysis to decompose functional images of patients with AD in orthogonal ensembles of brain regions with maximal metabolic covariance. Three principal components explained 38% of the total variance in a large sample of FDG-PET images obtained in 225 AD patients. One functional ensemble (PC2) included limbic structures from Papez's circuit (medial temporal regions, posterior and anterior cingulate cortex, thalamus); its disruption in AD patients was related to episodic memory impairment. Another principal component (PC1) illustrated major metabolic variance in posterior cerebral cortices, and patients' scores were correlated to instrumental functions (language and visuospatial abilities). PC3 comprised frontal, parietal, temporal and posteromedial (posterior cingulate and precuneus) cortices, and patients' scores were related to executive dysfunction and global cognitive impairment. The three main metabolic covariance networks converged in the posterior cingulate area that showed complex relationships with medial temporal structures within each PC. Individual AD scores were distributed as a continuum along PC axes: an individual combination of scores would determine specific clinical symptoms in each patient.