The aim of this study is to identify and validate copy number aberrations in early-stage primary breast tumors associated with bone or non-bone metastasis. Whole-genome molecular inversion probe arrays were used to evaluate copy number imbalances (CNIs) in breast tumors from 960 early-stage patients with information about site of metastasis. The CoxBoost algorithm was used to select metastasis site-related CNIs and to fit a Cox proportional hazards model. Gains at 1q41 and 1q42.12 and losses at 1p13.3, 8p22, and Xp11.3 were significantly associated with bone metastasis. Gains at 2p11.2, 3q21.3-22.2, 3q27.1, 10q23.1, and 14q13.2-3 and loss at 7q21.11 were associated with non-bone metastasis. To examine the joint effect of CNIs and clinical predictors, patients were stratified into three risk groups (low, intermediate, and high) based on the sum of predicted linear hazard ratios. For bone metastasis, the hazard (95 % confidence interval) for the low-risk group was 0.32 (0.11-0.92) compared to the intermediate-risk group and 2.99 (1.74-5.11) for the high-risk group. For non-bone metastasis, the hazard for the low-risk group was 0.34 (0.17-0.66) and 2.33 (1.59-3.43) for the high-risk group. The prognostic value of loss at 8p22 for bone metastasis and gains at 10q23.1 for non-bone metastasis, and gain at 11q13.5 for both bone and non-bone metastases were externally validated in 335 breast tumors pooled from four independent cohorts. Distinct CNIs are independently associated with bone and non-bone metastasis for early-stage breast cancer patients across cohorts. These data warrant consideration for tailoring surveillance and management of metastasis risk.