Atrial fibrillation (AF) is a morbid and heritable arrhythmia. Over 35 genes have been reported to underlie AF, most of which were described in small candidate gene association studies. Replication remains lacking for most, and therefore the contribution of coding variation to AF susceptibility remains poorly understood. We examined whole exome sequencing data in a large community-based sample of 1,734 individuals with and 9,423 without AF from the Framingham Heart Study, Cardiovascular Health Study, Atherosclerosis Risk in Communities Study, and NHLBI-GO Exome Sequencing Project and meta-analyzed the results. We also examined whether genetic variation was enriched in suspected AF genes (N = 37) in AF cases versus controls. The mean age ranged from 59 to 73 years; 8,656 (78%) were of European ancestry. None of the 99,404 common variants evaluated was significantly associated after adjusting for multiple testing. Among the most significantly associated variants was a common (allele frequency = 86%) missense variant in SYNPO2L (rs3812629, p.Pro707Leu, [odds ratio 1.27, 95% confidence interval 1.13-1.43, P = 6.6x10-5]) which lies at a known AF susceptibility locus and is in linkage disequilibrium with a top marker from prior analyses at the locus. We did not observe significant associations between rare variants and AF in gene-based tests. Individuals with AF did not display any statistically significant enrichment for common or rare coding variation in previously implicated AF genes. In conclusion, we did not observe associations between coding genetic variants and AF, suggesting that large-effect coding variation is not the predominant mechanism underlying AF. A coding variant in SYNPO2L requires further evaluation to determine whether it is causally related to AF. Efforts to identify biologically meaningful coding variation underlying AF may require large sample sizes or populations enriched for large genetic effects.