Experimental implementations of dictator games are found to differ in terms of their underlying strategic incentives. We explore this discovery in two separate directions. Theoretically, assuming identical other-regarding preferences, we show that the two most widely used protocols can generate strongly contrasting rational-choice predictions, from which different interpretations of dictator giving arise. Experimentally, a tailor-made experiment reveals significant differences between the two protocols but rejects full rationality as a satisfactory explanatory theory. Our findings indicate that several previously drawn conclusions regarding other-regarding preferences among humans distinguished by social class, gender, generation, nationality, etc. may be more ambiguous than hitherto believed.