Relational reasoning is central to much of human-unique cognition including artistic metaphor, scientific analogy. Whilemuch research has addressed the process of relational reasoning, the conditions under which relational reasoning is en-gaged in at all remains under-explored.This work examines the relationship between dimensions on which stimuli vary and the likelihood that these stimuli willbe processed relationally by adults. We use a modified relational-match-to-sample paradigm: One of the two choicescontains a relational match with the target, the other contains a partial object match. Changing dimensions on which thestimuli vary dramatically effects the likelihood that adults process them relationally (i.e. make relational matches) - from56% when stimuli vary on shape and color to 98% when stimuli vary on size alone. This is despite the relational contentof the task remaining identical throughout.We discuss implications of these results for designing stimuli, and for theories of relational reasoning generally.

Relational Match to Sample (RMTS) is a common test of relational reasoning involving matching cards based onthe relations “same” and “different”. Children below the age of five fail RMTS, even with corrective feedback. Given thatsuccess on RMTS depends on the ability to represent and compare ”same” and ”different”, such failure has been interpreted asindicative of the absence of these abilities (Penn, Holyoak & Povinelli, 2008; Hochmann, Mody & Carey, 2016).In the current studies three, four and five-year-old children were provided explicit instructions on RMTS. Results showsuccess in all groups, including three-year-olds - two years earlier than previous work. This suggests the ability to representand compare ”same” and ”different” emerges significantly earlier than spontaneous success on RMTS, undermining previousinterpretations. More generally, this work begins to explore the nature of the development which allows existing relationalreasoning capacities to be spontaneously deployed in RMTS.