A marker of Alzheimer's disease (AD) that can accurately diagnose disease at the earliest stage would significantly support efforts to develop treatments for early intervention. We have sought to determine the sensitivity and specificity of peripheral blood gene expression as a diagnostic marker of AD using data generated on HT-12v3 BeadChips. We first developed an AD diagnostic classifier in a training cohort of 78 AD and 78 control blood samples and then tested its performance in a validation group of 26 AD and 26 control and 118 mild cognitive impairment (MCI) subjects who were likely to have an AD-endpoint. A 48 gene classifier achieved an accuracy of 75% in the AD and control validation group. Comparisons were made with a classifier developed using structural MRI measures, where both measures were available in the same individuals. In AD and control subjects, the gene expression classifier achieved an accuracy of 70% compared to 85% using MRI. Bootstrapping validation produced expression and MRI classifiers with mean accuracies of 76% and 82%, respectively, demonstrating better concordance between these two classifiers than achieved in a single validation population. We conclude there is potential for blood expression to be a marker for AD. The classifier also predicts a large number of people with MCI, who are likely to develop AD, are more AD-like than normal with 76% of subjects classified as AD rather than control. Many of these people do not have overt brain atrophy, which is known to emerge around the time of AD diagnosis, suggesting the expression classifier may detect AD earlier in the prodromal phase. However, we accept these results could also represent a marker of diseases sharing common etiology.