Various classification methods have been used to predict the class of tissue samples based on gene expression data. Prediction Analysis for Microarrays (PAM) is one of the top classifiers that has been extensively used for cancer classification. In this paper a novel method of combining expression data from gene pairs is used to improve the overall accuracy of PAM. Recent studies suggest that deregulation of pathways, rather than individual genes, may be critical in triggering carcinogenesis. The pathway deregulation is often caused by the simultaneous deregulation of more than one genes in the pathway. Robust gene pair combinations may exploit these underlying bio-molecular reactions to provide better biomarkers for cancer, as compared to single genes. In this work, we used gene pair combinations, called doublets, to improve the accuracy of PAM. We validated the proposed approach with nine cancer datasets. The accuracy of PAM, using these doublets, increased consistently across these datasets, in some cases with a significant margin (13%).