OBJECTIVE : To assess clinical pathological characteristics and outcome of triple-negative breast cancers (TNBC) by androgen receptor (AR) protein expression. METHODS : We retrospectively evaluated AR by immunohistochemistry on core-needle biopsy, (CNB) and residual disease (RD) in a consecutive institutional series of TNBC patients treated with neo-adjuvant chemotherapy (NACT) between 2000 and 2017. We investigated univariate associations between AR-expression on CNB (using different cut-offs), clinical pathological variables, and pathologic complete response (pCR). Next, we used multiple correspondence analysis (MCA) to investigate the relationships between AR on CNB and standard clinical and pathological variables, including stromal tumor infiltrating lymphocytes (sTILs). Finally, we investigated the prognostic value of AR-expression on CNB and RD using the Fine and Gray model. RESULTS : We included 71 patients; median follow-up was 6.7 years. Considering the ≥ 1% cut-off, AR was present in 32% on the CNB and 14% on RD. AR-low (1-34% positive tumor cells) patients were associated with younger (premenopausal) age and AR-high (≥ 34% positive tumor cells) with older (postmenopausal) age. AR on CNB did not correlate with other features nor was it predictive for pCR or prognostic for metastatic outcome, regardless of the used cut-off. The MCA suggested that body mass index (BMI) affects the predictive role of AR-low and -high for pCR differently. AR-loss on RD was prognostic for a better 5-year distant disease-free survival (DDFS) as compared to RD with retained AR-expression (61.6% (95% CI 44.26-79.14) and 25.0% (95% CI 3.94-87.21), respectively; p = 0.01). CONCLUSIONS : Low and high AR-expression on CNB of TNBC were correlated with age and menopausal status but qualitative AR was not predictive for pCR. AR-loss on RD was prognostic for DDFS in TNBC patients treated with NACT.