The power of human reasoning and learning seems to rely on the ability to dynamically bind representations of objects into relational roles. Computational considerations imply that the complexity of relational reasoning will be bounded due to working-memory limits. I will describe recent research on analogical reasoning and related tasks in which we have been exploring the nature and limits of relational reasoning. Evidence from behavioral, neuropsychological, and functional-imaging studies suggest that specific subregions of human prefrontal cortex provide the neural substrate for relational reasoning.