Therapeutic drug monitoring (TDM) of β-lactam antibiotics is increasingly used to overcome rising antimicrobial resistance and improve antibiotic exposure. However, there is little guidance on target amoxicillin plasma concentrations. We aimed to define these by evaluating associations between amoxicillin concentrations and clinical outcomes. This single-centre prospective cohort study enrolled severely ill and/or immunosuppressed adult patients receiving amoxicillin for suspected or confirmed bacterial infection. TDM with ≥1 intermediate and ≥1 trough level was performed 24 h after therapy initiation. Primary and secondary outcomes were incidence of adverse events (AEs) and clinical failure through Day 30, respectively. A total of 156 patients were included. Important variations were observed both for intermediate (mean 13 mg/L, S.D. 13) and trough (mean 7 mg/L, S.D. 9) amoxicillin levels. Of 111 patients, 33 (30%) had trough levels below the non-species-related breakpoint (2 mg/L). AEs occurred in 27/156 patients (17%); no intermediate- or trough-level threshold predicting toxicity could be established. Patients with the highest-quartile trough levels (9.07-51.5 mg/L) did not experience significantly increased AEs [6/28 (21%) vs. 13/83 (16%); P = 0.6]. Nearly one-third (48/156; 31%) experienced clinical failure; low trough levels did not correlate with failure. There were few amoxicillin AEs yet a relatively high incidence of clinical failure. While no toxicity threshold could be established, the absence of increased AEs among patients with the highest trough concentrations suggests that trough levels up to 40 mg/L may be safe, at least for limited durations. Larger trials must further define optimal amoxicillin concentrations. [ClinicalTrials.gov ID: NCT03790631].