Summary Background Primary results of the HORIZONS-AMI trial have been previously reported. In this final report, we aimed to assess 3-year outcomes. Methods HORIZONS-AMI was a prospective, open-label, randomised trial undertaken at 123 institutions in 11 countries. Patients aged 18 years or older were eligible for enrolment if they had ST-segment elevation myocardial infarction (STEMI), presented within 12 h after onset of symptoms, and were undergoing primary percutaneous coronary intervention. By use of a computerised interactive voice response system, we randomly allocated patients 1:1 to receive bivalirudin or heparin plus a glycoprotein IIb/IIIa inhibitor (GPI; pharmacological randomisation; stratified by previous and expected drug use and study site) and, if eligible, randomly allocated 3:1 to receive a paclitaxel-eluting stent or a bare metal stent (stent randomisation; stratified by pharmacological group assignment, diabetes mellitus status, lesion length, and study site). We produced Kaplan-Meier estimates of major adverse cardiovascular events at 3 years by intention to treat. This study is registered with ClinicalTrials.gov , number NCT00433966. Findings Compared with 1802 patients allocated to receive heparin plus a GPI, 1800 patients allocated to bivalirudin monotherapy had lower rates of all-cause mortality (5·9% vs 7·7%, difference −1·9% −3·5 to −0·2, HR 0·75 0·58–0·97; p=0·03), cardiac mortality (2·9% vs 5·1%, −2·2% −3·5 to −0·9, 0·56 0·40–0·80; p=0·001), reinfarction (6·2% vs 8·2%, −1·9% −3·7 to −0·2, 0·76 0·59–0·99; p=0·04), and major bleeding not related to bypass graft surgery (6·9% vs 10·5%, −3·6% −5·5 to −1·7, 0·64 0·51–0·80; p=0·0001) at 3 years, with no significant differences in ischaemia-driven target vessel revascularisation, stent thrombosis, or composite adverse events. Compared with 749 patients who received a bare-metal stent, 2257 patients who received a paclitaxel-eluting stent had lower rates of ischaemia-driven target lesion revascularisation (9·4% vs 15·1%, −5·7% −8·6 to −2·7, 0·60 0·48–0·76; p<0·0001) after 3 years, with no significant differences in the rates of death, reinfarction, stroke or stent thrombosis. Stent thrombosis was high (≥4·5%) in both groups. Interpretation The effectiveness and safety of bivalirudin monotherapy and paclitaxel-eluting stenting are sustained at 3 years for patients with STEMI undergoing primary percutaneous coronary intervention. Funding Boston Scientific and The Medicines Company.