In gene expression analyses using a high-density oligonucleotide array in a rat ischemia model, two comparison methods, "pair-wise comparison" and "sample average comparison", were evaluated based on statistical methods. The reliability of the elements screened with a 1.2 to 10-fold threshold was also evaluated. In pair-wise comparisons, most of the elements were significantly independent of the threshold value, with the percentage of significant elements remaining above 95%, when screened at 2.5-fold or higher threshold value. Pair-wise comparison structurally provided strict screening, which resulted in genes that were not selected despite significant alterations in expression. Screening by "sample average comparison" resulted in elements with low probability of significance, which suggested the necessity for increasing the reliability by additional statistical analyses after screening. When genes with altered expression were screened using an oligonucleotide array, marked differences in the numbers and reliability were proved to exist among elements screened by each sample comparison method.