Most individuals fail the selection task, selecting P and Q cases, when they have to test descriptive rules of the form ``If P, then Q''. But they solve it, selecting P and not-Q cases, when they have to test deontic rules of the form ``If P, then must Q''. According to relevance theory, linguistic comprehension processes determine intuitions of relevance that, in turn, determine case selections in both descriptive and deontic problems. We tested the relevance theory predictions in a within-participants experiment. The results showed that the same rule, regardless of whether it is tested descriptively or deontically, can be made to yield more P and Q selections or more P and not-Q selections. We conclude that the selection task does not provide a tool to test general claims about human reasoning. (C) 2001 Elsevier Science B.V. All rights reserved.