Statistical comparisons of [(18)F]FDG PET scans between healthy subjects and patients with Alzheimer's disease (AD) or amnestic mild cognitive impairment (aMCI) using Statistical Parametric Mapping (SPM) usually require normalization of regional tracer uptake via ROIs defined using additional software. Here, we validate a simple SPM-based method for count normalization. FDG PET scans of 21 mild, 15 very mild AD, 11 aMCI patients and 15 age-matched controls were analyzed. First, we obtained relative increases in the whole patient sample compared to controls (i.e. areas relatively preserved in patients) with proportional scaling to the cerebral global mean (CGM). Next, average absolute counts within the cluster with the highest t-value were extracted. Statistical comparisons of controls versus three patients groups were then performed using count normalization to CGM, sensorimotor cortex (SMC) as standard, and to the cluster-derived counts. Compared to controls, relative metabolism in aMCI patients was reduced by 15%, 20%, and 23% after normalization to CGM, SMC, and cluster-derived counts, respectively, and 11%, 21%, and 25% in mild AD patients. Logistic regression analyses based on normalized values extracted from AD-typical regions showed that the metabolic values obtained using CGM, SMC, and cluster normalization correctly classified 81%, 89% and 92% of aMCI and controls; classification accuracies for AD groups (very mild and mild) were 91%, 97%, and 100%. The proposed algorithm of fully SPM-based count normalization allows for a substantial increase of statistical power in detecting very early AD-associated hypometabolism, and very high accuracy in discriminating mild AD and aMCI from healthy aging.