An ideal diagnostic test for Alzheimer's disease (AD) should be non-invasive and easily applicable. Thus, there is a clear need to search for biomarkers in blood. In the present study, we have used multivariate data analysis [support vector machine (SVM)] to investigate whether a blood-based biomarker panel allows discrimination between AD patients and healthy controls at the individual level. We collected a total of 155 serum samples from individuals with early AD and age-matched healthy controls and measured serum levels of 24 markers involved in several biological pathways by ELISA. The dataset was randomly split into a training set for predictor discovery and classification training and a test set for class prediction of blinded samples (3:1 ratio) to evaluate the chosen predictors and parameters. After selection of a feature group of the three most discriminative parameters (cortisol, von Willebrand factor, oxidized LDL antibodies) in the training set, the application of SVM to the training/independent test dataset resulted in an 81.7%/87.1% correct classification for AD and control subjects. In conclusion, we identified a panel of three blood markers, which allowed SVM-based distinguishing of AD patients from healthy controls on a single-subject classification level with clinically relevant accuracy and validity. Blood-based biomarkers might have utility in AD diagnostics as screening tool before further classification with CSF biomarkers and imaging. Future studies should examine whether blood-based biomarkers may also be useful to differentiate AD patients from other dementias.