Background— Safety and efficacy limit currently available atrial fibrillation (AF) therapies. We hypothesized that atrial gene transfer would allow focal manipulation of atrial electrophysiology and, by eliminating reentry, would prevent AF. Methods and Results— In a porcine AF model, we compared control animals to animals receiving adenovirus that encoded KCNH2-G628S, a dominant negative mutant of the I Kr potassium channel α-subunit (G628S animals). After epicardial atrial gene transfer and pacemaker implantation for burst atrial pacing, animals were evaluated daily for cardiac rhythm. Electrophysiological and molecular studies were performed at baseline and when animals were euthanized on either postoperative day 7 or 21. By day 10, none of the control animals and all of the G628S animals were in sinus rhythm. After day 10, the percentage of G628S animals in sinus rhythm gradually declined until all animals were in AF by day 21. The relative risk of AF throughout the study was 0.44 (95% confidence interval 0.33 to 0.59, P <0.01) among the G628S group versus controls. Atrial monophasic action potential was considerably longer in G628S animals than in controls at day 7, and KCNH2 protein levels were 61% higher in the G628S group than in control animals ( P <0.01). Loss of gene expression at day 21 correlated with loss of action potential prolongation and therapeutic efficacy. Conclusions— Gene therapy with KCNH2-G628S eliminated AF by prolonging atrial action potential duration. The effect duration correlated with transgene expression.