In this report, we address the following questions: What aspects of information do preservice elementary teachers rely on when performing inductive reasoning? What contexts enable them to perceive the inherent invariant relationships from a finite sample and, thus, formulate viable generalizations? To what extent are they able to justify inductive results noninductively? Our responses are based mostly on inductors' ability to perceive similarity between compared entities that they compute over numerical and figural cues. We tested a model of similarity and examined predictions of the model in the case of two problem tasks. In the model, we assume that both numerical and figural cues contribute to similarity leading to induction. An analysis of the induction processes of the 42 participants suggests that even if relationships among numerical values have had greater contribution to similarity than did figural ones, those who induced figurally acquired a better understanding of the generalizations they constructed. (Contains 3 tables.) [For complete proceedings, see ED500860.]