Background: Staphylococcus aureus persistent bacteraemia is only vaguely defined and the effect of different durations of bacteraemia on mortality is not well established. Our primary aim was to analyse mortality according to duration of bacteraemia and to derive a clinically relevant definition for persistent bacteraemia. Methods: We did a secondary analysis of a prospective observational cohort study at 17 European centres (nine in the UK, six in Spain, and two in Germany), with recruitment between Jan 1, 2013, and April 30, 2015. Adult patients who were consecutively hospitalised with monomicrobial S aureus bacteraemia were included. Patients were excluded if no follow-up blood culture was taken, if the first follow-up blood-culture was after 7 days, or if active antibiotic therapy was started more than 3 days after first blood culture. The primary outcome was 90-day mortality. Univariable and time-dependent multivariable Cox regression analysis were used to assess predictors of mortality. Duration of bacteraemia was defined as bacteraemic days under active antibiotic therapy counting the first day as day 1. Findings: Of 1588 individuals assessed for eligibility, 987 were included (median age 65 years [IQR 51–75]; 625 [63%] male). Death within 90 days occurred in 273 (28%) patients. Patients with more than 1 day of bacteraemia (315 [32%]) had higher Charlson comorbidity index and sequential organ failure assessment scores and a longer interval from first symptom to first blood culture. Crude 90-day mortality increased from 22% (148 of 672) with 1 day of bacteraemia, to 39% (85 of 218) with 2–4 days, 43% (30 of 69) with 5–7 days, and 36% (10 of 28) with more than 7 days of bacteraemia. Metastatic infections developed in 39 (6%) of 672 patients with 1 day of bacteraemia versus 40 (13%) of 315 patients if bacteraemia lasted for at least 2 days. The second day of bacteraemia had the highest HR and earliest cutoff significantly associated with mortality (adjusted hazard ratio 1·93, 95% CI 1·51–2·46; p<0·0001). Interpretation: We suggest redefining the cutoff duration for persistent bacteraemia as 2 days or more despite active antibiotic therapy. Our results favour follow-up blood cultures after 24 h for early identification of all patients with increased risk of death and metastatic infection. Funding: None.