One of the reasons why second-order correlation learning has often been overlooked in the literature is because it was previously assumed that two memory retrieval cues become associated because they are both present at the time of the target episode, which was the case in previous research that examined infants’ ability to encode clusters of correlated attributes (e.g., Younger & Cohen, 1986). However, it is not necessary that two cues have temporal contiguity for associative learning to occur as demonstrated in trace conditioning studies in which the conditioned stimulus is not physically present when the unconditioned stimulus is presented (Pavlov, 1927). This idea has been illustrated by Dwyer, Mackintosh, and Boakes (1998), who showed that rats can associate two cues that were not present simultaneously but that were simultaneously activated in memory.