Purpose To investigate the distribution of plasma apixaban levels and their relationships with clinical outcomes in elderly patients with atrial fibrillation (AF). Method The J-ELD AF Registry is a multicenter prospective observational study of Japanese non-valvular AF patients aged ≥75 years taking an on-label dose of apixaban (3015 patients from 110 institutions). Among them, plasma apixaban levels at trough were estimated by anti-Xa assay (Api-AXA) in 943 patients. Patients with standard (5 mg bid; n  = 431) and reduced (2.5 mg bid; n  = 512) dose were further divided into two groups with low and high Api-AXA levels (boundary: median value). Results The incidence rates (per 100 person-years) of events in the low- and high-Api-AXA groups were as follows: 1.48 and 1.99 (log-rank test, P  = 0.695) for stroke or systemic embolism, 0.98 and 1.49 ( P  = 0.652) for bleeding requiring hospitalization, and 0.49 and 0.99 ( P  = 0.565) for total deaths in patients with standard dose, versus 0.84 and 1.68 ( P  = 0.414), 0.42 and 4.64 ( P  = 0.004), and 2.52 and 6.65 ( P  = 0.035) in patients with a reduced dose, respectively. In multivariate Cox regression analysis among patients with a reduced dose, a high Api-AXA level was independently associated with bleeding requiring hospitalization (HR 12.12, 95% CI: 1.56–94.22) and nonsignificantly with total deaths. Conclusions A high trough apixaban level in patients indicated for standard dose was not associated with adverse events, while a high apixaban level in patients indicated for a reduced dose was associated with bleeding requiring hospitalization.