The standard approach posits that analogical inferences are generated by copying unmapped base relations, substituting base entities by their corresponding target ones, and generating slots for unmapped base entities. Contra this account, results from Experiment 1 revealed that analogical inferences seldom include relations that resemble the base relation from which they were derived. Most of the inferences, however, could be categorized as exemplars of a schema-governed category capable of characterizing the base information to be projected. To gather further precision about the criteria that guide inference generation, in Experiment 2 we showed that analogical inferences tend to match the base information from which they are derived in values of salient dimensions of the relational category to which they belonged. Our results suggest that the relational constructs employed in modeling analogical inference should move beyond one-term multiplace predicates so as to include more complex relational structures.