A new word-learning phenomenon is demonstrated and a new word-learning principle is proposed to account for it. In Study 1, 60 3-year-olds were shown a pair of objects and heard a novel label used repeatedly for one, but not for the other. In a forced-choice test of generalization of the label, the latter object was selected less often by the children than one that had not been present during training. This so-called Nominal Passover Effect was the same whether the speaker had completely ignored the comparison object during training or had referred to it with pronouns. The performance of a no-word control group (N = 24) indicated that the effect was not due to a preference for the less exposed of the two choice objects. The effect is consistent with the Exhaustive Reference Principle, which stipulates that whenever a new generic word is used to name something, expect it to be extended to all entities in a situation that the speaker perceives and believes to be exemplars of the name. In Study 2 (N = 48), the Nominal Passover Effect was replicated with 3 new sets of objects and with training language that contained only indefinite forms of reference. The passover experience was often sufficient to counteract children's tendency to generalize a novel label on the basis of perceptual similarity. The passover effect was not evident in free-choice name generalization tests in either study.