Reasoning based on relations is essential for human learning and thinking. However, how this ability is acquired remains unclear. Two accounts offer different views: whereas one suggests that relational reasoning is related to cognitive development, the other views it as a learned bias. We conducted two experiments to investigate whether relational reasoning is a learned bias. Experiment 1 tested adults on the Relational Match-to-Sample task. The results revealed that a significant proportion of adults failed to engage in the expected relational reasoning; instead, they relied on the object similarity. Scores on Raven Advanced Progressive Matrices (APM-12) suggest that the preference for similarity is not tied to general cognitive ability. Experiment 2 tested whether similarity-based reasoners can learn relational bias when prompted to generate explanations. The results showed that participants who primarily generated relational explanations successfully learned relational bias. Taken together, this study suggests that relational reasoning is a learned bias.