Introduction: Direct oral anticoagulants (DOAC) are guideline-recommended over warfarin for stroke prevention in atrial fibrillation (AF). However, patients who are DOAC eligible are commonly maintained on warfarin. We sought to evaluate bleeding risk and prediction while on DOAC treatment (both for warfarin-naïve and -experienced patients) as compared to warfarin. Methods: We performed a retrospective cohort study using data from the Veteran Affairs health care system. We included patients with a prescription for warfarin and/or DOAC from 10/1/2010 to 9/30/2017 with an AF encounter in the 90 days prior to 30 days after prescription. We categorized DOAC treated patients as warfarin-naïve or -experienced and performed an on-treatment analysis to determine bleeding incidence and HAS-BLED score discrimination. In adjusted analyses, we compared risk of bleeding while treated with DOAC (both for warfarin-naïve and -experienced patients) to warfarin. Results: The analysis cohort included 99,143 patients treated with warfarin (71±10 years, HAS-BLED 2.6±1.2) and 73,732 and 26,760 patients treated with DOAC who were warfarin-naïve (74±10 years, HAS-BLED 2.4±1.0) and -experienced (71±9 years, HAS-BLED 2.8±1.1), respectively. DOAC patients with warfarin experience had more prior bleeds (DOAC, warfarin-experienced: 11.9%; DOAC, warfarin-naïve: 4.5%; warfarin: 6.2%; p<0.001 for both). Risk of intracranial bleeding was substantially lower while on DOAC treatment (both for warfarin-naïve and -experienced patients) as compared to warfarin ( Table ). HAS-BLED discrimination for bleeding outcomes, intracranial or any bleeding, was modest ( Table ). Conclusion: DOAC treatment had a favorable safety profile compared to warfarin treatment, even for DOAC treated patients with warfarin-experience who had more prior bleeds. These data argue against maintaining DOAC eligible patients on warfarin therapy regardless of HAS-BLED score.