Purpose The influence of the androgen receptor (AR) CAG repeat polymorphism on breast cancer is controversial. We investigated the combined effects of CAG repeat length and estrogen receptor (ER) status on prognosis in 355 postmenopausal women with primary breast cancer. Methods CAG repeat length was determined by the HUMARA test. Relapse-free survival (RFS) and overall survival (OS) according to the X-weighted CAG repeat biallelic mean (XWBM) were investigated by univariate and multivariate analysis. Results XWBM was not associated with RFS or OS, but a significant interaction between XWBM and ER status (p = 0.002) was found for OS. ER-negative patients with median XWBM <20 showed lower OS than ER-negative/XWBM ≥20 patients (HR = 0.270; 95% Cl: 0.073-0.999). ER-negative/XWBM <20 patients also had significantly lower OS than ER-positive women, irrespective of CAG repeat length (p<0.001). Accordingly, estimated OS was lowest in ER-negative patients with XWBM <20 (OS: 0.63, 95% CI: 0.41-0.79) and highest in ER-positive patients with XWBM <20 (OS: 0.95, 95% CI: 0.90-0.97). Conclusions Our data suggest that short CAG repeat length is associated with increased risk of death in ER-negative disease but is related to better survival when ER is expressed. These findings are in agreement with the hypothesis that AR may stimulate or inhibit breast cancer growth depending on ER status, AR transactivation, and the endocrine-metabolic environment of breast tumors. Evaluation of CAG repeat length together with ER status could help improve the estimation of the risk of death, with possible implications for the optimization of standard breast cancer treatment and implementation of prevention strategies.