A fundamental property of human memory is that repetition enhances memory. Peterson and Mulligan (2012) recently documented a surprising "negative repetition effect," in which participants who studied a list of cue-target pairs twice recalled fewer targets than a group who studied the pairs only once. Words within a pair rhymed, and across pairs, the target words were drawn from a small set of categories. In the repetition condition, the pairs were initially presented in a random order and then presented a 2nd time blocked by the category of the target words. In the single presentation condition, the pairs were presented only in the blocked order. Participants in the former condition recalled fewer target words on a free recall test despite having seen the word pairs twice (the negative repetition effect). This phenomenon is explored in a series of 5 experiments assessing 3 theoretical accounts of the effect. The experiments demonstrate that the negative repetition effect generalizes over multiple encoding conditions (reading and generative encoding), over different memory tests (free and cued recall), and over delay (5 min and 2 days). The results argue against a retrieval account and a levels-of-processing account but are consistent with the item-specific--relational account, the account upon which the effect was initially predicated.