Background: Atrial fibrillation (AF) is associated with high risk of mild cognitive impairment (MCI) and dementia. However, feasible and simple instruments that facilitates regular assessment of cognitive status in AF patients remain underdeveloped. Methods: Cognitive function of 136 AF patients was first evaluated using T-MoCA. Cognitive function of 101 patients was then assessed through in-person interview by physicians who are blinded to telephone interview results, using both Clinical Dementia Rating (CDR) and Mini-Mental Status Evaluation (MMSE). Using CDR=0.5 as a reference standard, the ability of T-MoCA and MMSE to discriminate cognitive dysfunction, stratified by education level, was tested by receiver-operating curve (ROC) analysis. Net reclassification index was calculated for comparison between the performance of T-MoCA and MMSE. Results: Thirty-five MCI patients were identified as MCI using the criteria of CDR=0.5. The areas under the ROC curve of T-MoCA were 0.80 (0.71-0.89), 0.83 (0.71-0.95), and 0.85 (0.64-0.92) for all patients, patients with high educational level, and patients with low education level, respectively. The optimal threshold was achieved at 16/17 with a sensitivity of 85.7% and a specificity of 69.7% in overall patients, 15/16 with a sensitivity of 88.2% and a specificity of 64.5% in the low educational level patients, and 16/17 with a sensitivity of 77.8% and a specificity of 87.9% in the high educational level patients. Compared to the criterion MMSE≤27 and MMSE norms for the Chinese community elderly, stratified T-MoCA threshold improves correct classification by 23.7% (p=0.033) and 30.3% (p=0.020) respectively. Conclusion: T-MoCA is a feasible and effective method for MCI screening in patients with AF.