The expression profiling and molecular grouping of stomach cancers has been a challenging task because of their complexity and variation. We have analyzed gene expression profiles of 22 gastric cancer/nontumor mucosa couples using 14K cDNA microarray chips designed for gastric cancer analysis. Upon pairwise analysis of the individual couples at the false significance rate 0.91%, 79 and 398 genes were reported to be up-regulated and down-regulated in tumors, respectively. Tumors were clustered into two groups having high and low inflammatory infiltration, respectively. The latter consisted of three subgroups, including diffuse type carcinomas and intestinal types with distinct pathological characteristics of aggressive behavior. When the pooled tumor was hybridized against the pooled nontumor mucosa samples, more genes were detected to express differentially than those detected by the pairwise analysis at the same threshold level. However, they did not render satisfactory clustering of individual tumors. Our data showed that stomach cancers could be clustered effectively using stomach-specific microarrays and pairwise analysis of tumor/nontumor mucosa couples. It is suggested that the application of specific goal-oriented experimental designing would be advantageous for efficient analysis of expression profiles of such a complex disease as gastric cancer.