Gliomas are among the most lethal and debilitating cancers. Genetic testing is a rapidly evolving modality for cancer management. The advent of DNA microarrays enabled the utility of computational analyses in such management on a molecular basis. However, as current computational analyses remain insensitive to interactions between molecular features, they rarely postulate reasonable pathogenesis. The current study proposes a heuristic feature selection algorithm that identifies subsets of genes to almost perfectly classify glioma grades. The discretization technique in our method is a powerful tool against the tremendous data volume in DNA microarray. Instead of recognizing individual genetic features, the proposed algorithm helps identify specific gene subsets that play important roles in the pathogenesis of glioma.