PURPOSE: We tried to select and validate the candidate gene for the prognostic marker of breast cancer by comparing the analysis of copy number variation (CNV) between normal breast tissues and breast cancer tissues by performing array comparative genomic hybridization (CGH). METHODS: Array CGH was performed with using the fresh frozen tissues of 77 breast cancer patients. We selected the clones with more than a 20% frequency of gain or loss, and the clones with gain or loss in more than 2 consecutive clones. We finally selected the clones that were statistically significant on the survival analysis. We searched for the candidate gene that belonged to the candidate clones and we selected the final candidate gene that is assumed to be most related to the carcinogenesis of breast cancer by searching for information of the individual gene. We performed RT-PCR to validate the RNA expression of the final candidate gene with using the breast tissues of another 20 breast cancer patients. RESULTS: Eleven (10 in the gain group and 1 in the loss group) clones were finally selected as candidate clones. The significant CNVs with gain were found in the regions of 1q23.1, 1q41, 1q44, 5p15.33, 8q21.3, 15q26.3, 17q12 and 21q22.3 and the significant CNV with loss was found in 14q32.33. COL18A1 (21q22.3) was selected as the final candidate gene and the RT-PCR results revealed that the expression of COL18A1 was up-regulated in the cancer tissues of 18 of the other 20 (90%) breast cancer patients. CONCLUSION: We selected COL18A1 (21q22.3) as the candidate gene for the prognostic marker of breast cancer by comparing the analysis of CNVs from the array CGH. The RNA of COL18A1 was over-expressed in breast cancer tissue, as determined by RT-PCR.