Atrial fibrillation (AF) is the most common sustained clinical arrhythmia and is a problem of growing proportions. Recent studies have increased interest in fast-unbinding Na⁺-channel blockers like vernakalant (RSD1235) and ranolazine for AF therapy, but the mechanism of efficacy is poorly understood. To study how fast-unbinding I_Na-blockers affect AF, we developed realistic mathematical models of state-dependent Na⁺-channel block, using a lidocaine model as a prototype, and studied effects on simulated cholinergic AF in 2- and 3 dimensional atrial substrates. We then compared the results with in vivo effects of lidocaine on vagotonic AF in dogs. Lidocaine action was modeled with Hondeghem-Katzung modulated-receptor theory and maximum affinity for activated Na⁺-channels. Lidocaine produced frequency-dependent Na⁺-channel blocking and conduction slowing effects, and terminated AF in both 2- and 3-dimensional models with concentration-dependent efficacy (maximum ~89% at 60 µM). AF termination was not related to increases in wavelength, which tended to decrease with the drug, but rather to decreased source Na⁺-current in the face of large I_KACh-related sinks, leading to destabilization of primary-generator rotors and a great reduction in wavebreak, which caused primary-rotor annihilations in the absence of secondary rotors to resume generator activity. Lidocaine also reduced the variability and maximum values of the dominant frequency distribution during AF. Qualitatively similar results were obtained in vivo for lidocaine effects on vagal AF in dogs, with an efficacy of 86% at 2 mg/kg IV. These results provide new insights into the mechanisms by which rapidly-unbinding class I antiarrhythmic agents, a class including several novel compounds of considerable promise, terminate AF.