Abstract Background Identifying features associated with atrial fibrillation (AF) documentation could inform screening. This study used published data to describe differences in documented and estimated AF prevalence in general practices, and explored predictors of variations in AF prevalence. Methods Cross-sectional study of 7318 general practices in England. Descriptive and inferential statistics were undertaken. Multiple linear regression was used to model the difference between estimated AF and documented AF, adjusted for population, practice and practice performance variables. Results Documented AF prevalence was lower than estimated (− 0.55% 95% confidence intervals, −1.89, 2.99). The proportion of variability accounted for in the final regression model was 0.25. Factors positively associated with AF documentation (increase in difference between estimated and documented), were patients 65–74 years, 75 years +, Black or South Asian ethnicity, diabetes mellitus and practices in East and Midlands of England. Eight variables (female patients, deprivation score, heart failure and peripheral artery disease, total patients per practice, full-time GPs and nurses; and location in South of England) were negatively associated with AF documentation (reduction in difference). Conclusion Variations in AF documentation were predicted by several practice and population characteristics. Screening could target these sources of variation to decrease variation and improve AF documentation.