Background: Prediction of late metastasis is of clinical relevance in breast cancer. However, systematic genome-wide studies to identify genes associated with increased risk of metastasis 5 or more years after surgery are scarce. Methods: We examined the natural course of disease in three previously published cohorts (Mainz, Rotterdam, Transbig) including 766 node-negative breast cancer patients with gene array data who did not receive systemic chemotherapy in the adjuvant setting. We established a Cox regression based method adjusted for multiple testing that identified genes predicting late metastasis (5 or more years after surgery). Only those genes were accepted that showed similar results in all three cohorts. Metastasis-free survival (MFS) was analyzed with univariate and multivariate Cox regression. Results: We identified 9 genes [ABCC5 (Hazard Ratio (HR) 2.19, p=0.003), EDDM3B (HR 3.58, p=0.044), RAD23B (HR 0.37, p<0.001), XYLT2 (HR 2.19, p=0.027), DDX18 (HR 0.35, p=0.006), GPBP1L1 (HR 0.20, p=0.018), UBB (HR 9.73, p=0.025), RPS24 (HR 0.20, p=0.050), GPC1 (HR 2.36, p=0.013)] predicting late metastasis. These genes retained their independent prognostic significance after adjustment for established clinical factors (age, tumor size, grade, hormone receptor status, HER2) and biological motives like estrogen receptor, proliferation, B or T cells. These late-type genes are largely associated with resistance to hypoxia, apoptosis and DNA damage, suggesting that they might contribute to persistence of disseminated tumor cells. Conclusions: Genes associated with late metastasis offer a perspective to identify breast cancer patients suitable for additional and prolonged therapies.