Purpose: Anticoagulant therapy with warfarin reduces the rate of thromboembolic events but increases the risk of bleeding in individuals with nonvalvular atrial fibrillation. Our goal was to evaluate, by usability testing, a web-based atrial fibrillation decision support tool (AF-DST) that provides additional information on the risk-benefit tradeoff of anticoagulation based upon the patient's specific risk factors, which were pre-populated from the medical record. Methods: We performed extensive simulations of nine hypothetical outpatient visits involving eight physicians, ranging from interns to residents to attending physicians. Usability testing was employed to record positive and negative critical incidents in the simulations. Simulations were recorded and analyzed by two investigators. Clinician satisfaction with the tool was measured with the Computer System Usability Questionnaire (CSUQ), using a Likert scale scored 1 to 7, and a second questionnaire specific to the tool also using a Likert scale scored 1-7. We also tested subjects' knowledge concerning the risks and benefits of anticoagulation before and after administering the AF-DST. Results: Respondents' knowledge scores for stroke and bleeding risk increased from 38% before using the AF-DST to 62% after (p=0.29). The mean score on the CSUQ was 6, indicating high user satisfaction. The mean score of 6.5 on the second Likert scale indicated that users believed that the AF-DST helped them make more informed decisions. In usability testing debriefing, clinicians favored entering the risk factor data themselves to ensure accuracy rather than having it entered automatically from the electronic medical record. Most physicians said they would likely reconsider their treatment decision when the tool recommended another strategy (mean score: 5), but clinicians felt somewhat uncomfortable not understanding how the “black box” calculations were made. Conclusions: A patient-specific AF-DST can increase clinicians' knowledge regarding risks and benefits of anticoagulation in nonvalvular atrial fibrillation. Satisfaction with the web-based decision support tool was high, but clinicians preferred to enter the patient-specific risk factors to “manually” confirm the accuracy of the clinical data rather than having them extracted automatically from the electronic medical record.