BACKGROUND: Pembrolizumab showed durable antitumour activity and manageable safety in metastatic triple-negative breast cancer in the single-arm KEYNOTE-012 and KEYNOTE-086 trials. In this study, we compared pembrolizumab with chemotherapy for second-line or third-line treatment of patients with metastatic triple-negative breast cancer. METHODS: KEYNOTE-119 was a randomised, open-label, phase 3 trial done at 150 medical centres (academic medical centres, community cancer centres, and community hospitals) in 31 countries. Patients aged 18 years or older, with centrally confirmed metastatic triple-negative breast cancer, Eastern Cooperative Oncology Group performance status of 0 or 1, who had received one or two previous systemic treatments for metastatic disease, had progression on their most recent therapy, and had previous treatment with an anthracycline or taxane were eligible. Patients were randomly assigned (1:1) using a block method (block size of four) and an interactive voice-response system with integrated web-response to receive intravenous pembrolizumab 200 mg once every 3 weeks for 35 cycles (pembrolizumab group), or to single-drug chemotherapy per investigator's choice of capecitabine, eribulin, gemcitabine, or vinorelbine (60% enrolment cap for each; chemotherapy group). Randomisation was stratified by PD-L1 tumour status (positive [combined positive score (CPS) ≥1] vs negative [CPS <1]) and history of previous neoadjuvant or adjuvant treatment versus de-novo metastatic disease at initial diagnosis. Primary endpoints were overall survival in participants with a PD-L1 combined positive score (CPS) of 10 or more, those with a CPS of 1 or more, and all participants; superiority of pembrolizumab versus chemotherapy was tested in all participants only if shown in those with a CPS of one or more. The primary endpoint was analysed in the intention-to-treat population; safety was analysed in the all-subjects-as-treated population. This Article describes the final analysis of the trial, which is now completed. This trial is registered with ClinicalTrials.gov, number NCT02555657. FINDINGS: From Nov 25, 2015, to April 11, 2017, 1098 participants were assessed for eligibility and 622 (57%) were randomly assigned to receive either pembrolizumab (312 [50%]) or chemotherapy (310 [50%]). Median study follow-up was 31·4 months (IQR 27·8-34·4) for the pembrolizumab group and 31·5 months (27·8-34·6) for the chemotherapy group. Median overall survival in patients with a PD-L1 CPS of 10 or more was 12·7 months (95% CI 9·9-16·3) for the pembrolizumab group and 11·6 months (8·3-13·7) for the chemotherapy group (hazard ratio [HR] 0·78 [95% CI 0·57-1·06]; log-rank p=0·057). In participants with a CPS of 1 or more, median overall survival was 10·7 months (9·3-12·5) for the pembrolizumab group and 10·2 months (7·9-12·6) for the chemotherapy group (HR 0·86 [95% CI 0·69-1·06]; log-rank p=0·073). In the overall population, median overall survival was 9·9 months (95% CI 8·3-11·4) for the pembrolizumab group and 10·8 months (9·1-12·6) for the chemotherapy group (HR 0·97 [95% CI 0·82-1·15]). The most common grade 3-4 treatment-related adverse events were anaemia (three [1%] patients in the pembrolizumab group vs ten [3%] in the chemotherapy group), decreased white blood cells (one [<1%] vs 14 [5%]), decreased neutrophil count (one [<1%] vs 29 [10%]), and neutropenia (0 vs 39 [13%]). 61 (20%) patients in the pembrolizumab group and 58 (20%) patients in the chemotherapy group had serious adverse events. Three (<1%) of 601 participants had treatment-related adverse events that led to death (one [<1%] in the pembrolizumab group due to circulatory collapse; two [1%] in the chemotherapy group, one [<1%] due to pancytopenia and sepsis and one [<1%] haemothorax). INTERPRETATION: Pembrolizumab did not significantly improve overall survival in patients with previously treated metastatic triple-negative breast cancer versus chemotherapy. These findings might inform future research of pembrolizumab monotherapy for selected subpopulations of patients, specifically those with PD-L1-enriched tumours, and inform a combinatorial approach for the treatment of patients with metastatic triple-negative breast cancer. FUNDING: Merck Sharp & Dohme.