Evaluation of androgen assay results using a curated Hershberger database

•We used 39 reference chemicals with reproducible AR pathway effects in rodents to evaluate the ToxCast/Tox 21 AR model based on 11 in vitro assays.•Agreement was 66% (19/29), with ten inconclusive in vitro model chemicals; most discrepancies were explained by in vitro to in vivo extrapolation.•In vitro false negatives were low potency chemicals in vivo; lowest effect levels were above doses estimated from top in vitro tested concentration.•The AR model had 100% positive predictive value, i.e. no false positives and chemicals with conclusive results were consistently positive in vivo.•The AR model is a critical tool for rapidly screening chemicals and anchoring adverse effects in intact animals to an endocrine mode of action.