In attempting to understand cooperation, economists have used the methods of experimental economics to focus on spheres of human behavior in which humans display altruism, reciprocity, or other social preferences through giving and through punishment. Recent work has begun to examine whether allowing allocations in the negative domain, that is, allowing subjects to take (or steal) other subjects’ endowments, might affect participants’ behavior. If participants’ behavior is affected, then our understanding of experimental results generally, and social preferences speci cally, should be affected too (List 2007, Bardsley 2008). In this paper we propose an experimental variation on the Dictator Game with third-party punishment (Fehr and Fischbacher 2004b). We examine, first, a basic Dictator Game with third-party punishment, after which we introduce a treatment allowing the dictator to take from the receiver, in the knowledge that the third party could punish them. The results conflict. Many dictators choose the most self-interested option, while, when taking is introduced as an option for the dictator, third parties punish the most self-interested option more than in the baseline.