Estimated cardiorespiratory fitness (e-CRF) based on readily available clinical and self-reported data is a promising alternative to the costly traditional assessment of CRF using exercise equipment, but its role as a predictor for incident atrial fibrillation (AF) is unclear. This study included 10,126 participants (54.5% women, 35% African-American, mean age 63.2 years) from the Reasons for Geographic and Racial Differences in Stroke study who were free of AF at baseline. Baseline (2003 to 2007) e-CRF was determined using a previously validated nonexercise algorithm. Incident AF cases were identified at a follow-up examination by electrocardiogram and self-reported medical history of previous physician diagnosis. After a median follow-up of 9.4 years, 906 participants (8.9%) developed AF. In a multivariable logistic regression model adjusted for sociodemographics and baseline cardiovascular disease risk factors as well as incident coronary heart disease, heart failure, and stroke, each 1-metabolic equivalent of task increase in e-CRF was associated with a 5% lower risk of AF development (odds ratio [95% CI] 0.95 [0.92 to 0.99]; p = 0.0129). This association was stronger in women (OR [95% CI] 0.85 (0.79, 0.92) than in men (OR (95% CI) 0.88 (0.84, 0.93), interaction p value = 0.05. No significant interactions by age, race, history of cardiovascular disease, or physical limitations were observed. In conclusion, e-CRF using a nonexercise algorithm is a useful predictor of incident AF, which is consistent with previous reports using traditional CRF. This suggests that e-CRF using nonexercise algorithms may serve as a useful alternative to CRF measured by costly and time-consuming exercise testing.