Experiments 1, 2, and 3 examined whether, under what conditions, and at what age infants form second-order correlations. Experiment 4 was designed to examine how infants form second-order correlations. Specifically, the experiment examined if infants are able to form a unique representation of the familiarization bodies with both parts on them, despite only seeing the parts one at a time and never seeing the body with both parts on it simultaneously. We addressed this question by using the same basic methodology as Experiment 1 but used the same object bodies in the test trials that were presented during familiarization; that is, rather than testing infants with a novel object body, we used the same object body that infants had previously experienced. We hypothesized that if infants had represented the objects implied by the second-order correlation—that is, the body and the two parts connected to it—without seeing them previously, then they would treat them as familiar in the test phase and would not examine them extensively.