Three experiments demonstrate that association-specific repetition effects can be obtained for both newly formed and preexisting associations and that these effects are sensitive to modality of presentation. After studying a list of word pairs, participants were shown the original intact pairs and pairs formed by recombining the original pairs. In a lexical-decision task in which participants were asked to indicate whether both items were words, responses were faster to newly formed associations in the intact than in the recombined condition. This association-specific repetition priming effect was also observed for preexisting associations when a speeded relatedness judgment task was used. Both effects were found to be attenuated under cross-modal presentation. Finally, an explicit speeded recognition task revealed an associative effect that was not attenuated when modality was crossed for newly formed associations and was actually exaggerated for preexisting associations, suggesting that the repetition priming effects were not produced by conscious recollection. Results are discussed in terms of frameworks that are based either on perceptual representation systems or on a transfer-appropriate processing model.