Three experiments that investigated alternative accounts of repetition priming are reported. All 3 experiments used semantic comparison tasks and included trials in which each of the 2 words being compared had previously occurred on separate trials. In the re-pair match condition, the response required matched that on the 2 previous trials in which the words had occured. In there-pair mismatch condition, the response required was opposite to that on the previous trials containing the words forming the critical pair. In all 3 experiments, responses were faster and more accurate in the re-pair match condition than in the re-pair mismatch condition. Possible accounts of this effect within the frameworks of instance theory and of transfer appropriate processing are considered.