Most scholars believe that rumors spark ethnic violence around the world, yet political scientists have been slow to understand the dynamics of these rumors, including whether they emerge and survive as a matter of course or require certain preconditions. Because empirical observation cannot readily answer this question, we use agentbased modeling to explore whether rumor survival depends on selected group follower and leader characteristics. Our first set of experiments assumes one group. They show that violence-promoting rumors are not inevitable, nor are group leaders necessary for their propagation. But rumor survival peaks when leaders espouse extreme beliefs and interact frequently with their followers. Our second set of experiments assumes rival groups whose members interact on a limited basis. Some of the experiments confirm the one-group conclusions. Others reveal that when one group's leaders persist in advocating moderation, rumor propagation remains low in both groups. Finally, when rival leaders try to outbid each other, rumor survival increases markedly in both groups. By way of conclusion, we consider the policy implications of these findings.