A probabilistic test (FUMI theory) for GeneChip experiments has been proposed for selecting the genes which show significant differences in the gene expression levels between a single pair of treatment and control. This paper describes that the reliability of the judgment by the FUMI theory can be enhanced, when the selected genes are referred to biomolecular-functional networks of a commercial database. The genes judged as being differently expressed are grouped into a cluster in the biomolecular networks. It is also demonstrated that false positive genes have a trend in the networks to be isolated from each other, and also away from the clustered genes, since the false positive genes are randomly selected.