Breast cancer research is hampered by difficulties in obtaining and studying primary human breast tissue, and by the lack of in vivo preclinical models that reflect patient tumor biology accurately. To overcome these limitations, we propagated a cohort of human breast tumors grown in the epithelium-free mammary fat pad of severe combined immunodeficient (SCID)/Beige and nonobese diabetic (NOD)/SCID/IL-2γ-receptor null (NSG) mice under a series of transplant conditions. Both models yielded stably transplantable xenografts at comparably high rates (∼21% and ∼19%, respectively). Of the conditions tested, xenograft take rate was highest in the presence of a low-dose estradiol pellet. Overall, 32 stably transplantable xenograft lines were established, representing 25 unique patients. Most tumors yielding xenografts were “triple-negative” [estrogen receptor (ER)−progesterone receptor (PR)−HER2+; n = 19]. However, we established lines from 3 ER−PR−HER2+ tumors, one ER+PR−HER2−, one ER+PR+HER2−, and one “triple-positive” (ER+PR+HER2+) tumor. Serially passaged xenografts show biologic consistency with the tumor of origin, are phenotypically stable across multiple transplant generations at the histologic, transcriptomic, proteomic, and genomic levels, and show comparable treatment responses as those observed clinically. Xenografts representing 12 patients, including 2 ER+ lines, showed metastasis to the mouse lung. These models thus serve as a renewable, quality-controlled tissue resource for preclinical studies investigating treatment response and metastasis. Cancer Res; 73(15); 4885–97. ©2013 AACR.