Recently a new high-throughput biomarker discovery platform based on printed glycan arrays (PGA) has emerged. PGAs are similar to DNA arrays but contain deposits of various carbohy-drate structures (glycans) instead of spotted DNAs. PGA-based biomarker discovery for the early detection, diagnosis and prognosis of human malignancies is based on the response of the immune system as measured by the level of binding of anti-glycan antibodies from human serum to the glycans on the ar-ray. Since the PGA offer a multitude of markers which can have moderate individual diagnostic power they can be combined in order to achieve maximal classification precision assessed by the popular performance measure area under the ROC curve (AUC). This paper presents an empirical analysis of several combination approaches including those that are specifically designed to maximize the AUC and those that are not, such as Fisher Linear Discriminant, Support Vector Machines and Gen-eralized Linear Model. The analysis is performed on real-life PGA data from three pilot studies involving malignant mesothe-lioma, lung cancer and ovarian cancer.