Rivaroxaban exposure and patient characteristics may affect the rivaroxaban benefit–risk balance. This study aimed to quantify associations between model-predicted rivaroxaban exposure and patient characteristics and efficacy and safety outcomes in patients with non-valvular atrial fibrillation (NVAF), using data from the phase 3 ROCKET AF trial (NCT00403767). In ROCKET AF, 14,264 patients with NVAF were randomized to rivaroxaban (20 mg once daily [OD], or 15 mg OD if creatinine clearance was 30–49 mL/min) or dose-adjusted warfarin (median follow-up: 707 days); rivaroxaban plasma concentration was measured in a subset of 161 patients. In this post hoc exposure–response analysis, a multivariate Cox model was used to correlate individual predicted rivaroxaban exposures and patient characteristics with time-to-event efficacy and safety outcomes in 7061 and 7111 patients, respectively. There was no significant association between model-predicted rivaroxaban trough plasma concentration (Ctrough) and efficacy outcomes. Creatinine clearance and history of stroke were significantly associated with efficacy outcomes. Ctrough was significantly associated with the composite of major or non-major clinically relevant (NMCR) bleeding (hazard ratio [95th percentile vs. median]: 1.26 [95% confidence interval 1.13–1.40]) but not with major bleeding alone. The exposure–response relationship for major or NMCR bleeding was shallow with no clear threshold for an acceleration in risk. History of gastrointestinal bleeding had a greater influence on safety outcomes than Ctrough. These results support fixed rivaroxaban 15 mg and 20 mg OD dosages in NVAF. Therapeutic drug monitoring is unlikely to offer clinical benefits in this indication beyond evaluation of patient characteristics.