Drug-induced block of the cardiac rapid delayed rectifying potassium current (I Kr), carried by the human ether-a-go-go-related gene (hERG) channel, is the most common cause of acquired long QT syndrome. Indeed, some, but not all, drugs that block hERG channels cause fatal cardiac arrhythmias. However, there is no clear method to distinguish between drugs that cause deadly arrhythmias and those that are clinically safe. Here we propose a mechanism that could explain why certain clinically used hERG blockers are less proarrhythmic than others. We demonstrate that several drugs that block hERG channels, but have favorable cardiac safety profiles, also evoke another effect; they facilitate the hERG current amplitude in response to low-voltage depolarization. To investigate how hERG facilitation impacts cardiac safety, we develop computational models of I Kr block with and without this facilitation. We constrain the models using data from voltage clamp recordings of hERG block and facilitation by nifekalant, a safe class III antiarrhythmic agent. Human ventricular action potential simulations demonstrate the ability of nifekalant to suppress ectopic excitations, with or without facilitation. Without facilitation, excessive I Kr block evokes early afterdepolarizations, which cause lethal arrhythmias. When facilitation is introduced, early afterdepolarizations are prevented at the same degree of block. Facilitation appears to prevent early afterdepolarizations by increasing I Kr during the repolarization phase of action potentials. We empirically test this prediction in isolated rabbit ventricular myocytes and find that action potential prolongation with nifekalant is less likely to induce early afterdepolarization than action potential prolongation with dofetilide, a hERG channel blocker that does not induce facilitation. Our data suggest that hERG channel blockers that induce facilitation increase the repolarization reserve of cardiac myocytes, rendering them less likely to trigger lethal ventricular arrhythmias.