We tried to establish models that predict systemic recurrence in breast cancer by selecting marker clones with DNA copy number alterations (CNAs) using an array comparative genomic hybridization (CGH). Array CGH containing 4,044 human bacterial artificial chromosome clones was used to assess CNAs in 62 primary breast cancer tissues from 31 patients with systemic recurrence within 5 years after surgery and clinicopathologically well matched 31 patients who had no evidence of disease for at least 5years. Fourteen significant clones (11 clones showing gain and 3 showing loss) were identified by systemic recurrence-free survival (SRFS) analysis and 23 significant clones (17 clones showing gain and 6 showing loss) identified by chi(2) test and FDR test were selected as predictive markers of systemic breast cancer recurrence. The significant CNAs were found in the chromosomal regions of 5p15.33, 11q13.3, 15q26.3, 17q25.3, 18q23 and 21q22.3 with gain and 9p12, 11q24.1 and 14q32.33 with loss. We devised 2 prediction models for the systemic recurrence of breast cancer based on the 14 clones and the 23 clones, respectively. The survivals of the patients were significantly separated according to the scores from each model at the optimal cut off values in SRFS and overall survival analysis. We found candidate clones and genes of which CNAs were significantly associated with systemic recurrence of breast cancer. The devised prediction models with these clones were effective at differentiating the recurrence and nonrecurrence.