PAM50-defined breast cancer intrinsic subtypes and risk-of-relapse (ROR) scores are prognostic and predictive of endocrine therapy and some chemotherapy. We investigated the prognostic and predictive effect of PAM50 classifications by chemotherapy type. NCIC CTG MA.21 randomized 2,104 patients to doxorubicin, cyclophosphamide, and paclitaxel (AC/T); dose-intense cyclophosphamide, epirubicin, and flurouracil (CEF); or dose-dense, dose-intense epirubicin, cyclophosphamide, and paclitaxel (EC/T). Patients were ≤60 years, with node-positive or high-risk node-negative disease, with median 8-year follow-up. Intrinsic subtypes and ROR were determined from RNA extracted from formalin-fixed paraffin-embedded sections by the NanoString PAM50 test. Univariate effects on relapse-free survival (RFS) were assessed with stratified log-rank test; multivariate analyses utilized stratified Cox regression. Among 1094 cases completing PAM50 intrinsic subtyping, 27 % were classified as luminal A, 23 % luminal B, 18 % HER2E, and 32 % basal-like. CEF and EC/T were superior to AC/T (p = 0.01). Higher continuous ROR was multivariately associated with worse RFS (p = 0.03), although categorical ROR was neither prognostic nor predictive. Intrinsic subtypes had a significant multivariate prognostic effect on RFS (p = 0.002). Compared with luminal A, hazard ratios were luminal B = 1.48 (95 % CI 0.92-2.37); HER2E = 2.68 (95 % CI 1.60-4.48); and basal-like = 1.97 (95 % CI 1.10-3.53). Intrinsic subtypes were not predictive of treatment benefit (AC/T vs. EC/T + CEF); however, subgroup analysis indicated subtypes (non-luminal vs. luminal) was predictive of taxane benefit (EC/T vs. CEF; p = 0.05). Both NanoString PAM50 subtypes and continuous ROR had significant prognostic effects on RFS for breast cancer patients treated with CEF, EC/T, and AC/T. Non-luminal tumors differentially responded to EC/T (with taxane) over CEF.