We propose a general and predictive explanation of the Wason Selection Task (where subjects are asked to select evidence for testing a conditional "rule"). Our explanation is based on a reanalysis of the task, and on Relevance Theory. We argue that subjects' selections in all true versions of the Selection Task result from the following procedure. Subjects infer from the rule directly testable consequences. They infer them in their order of accessibility, and stop when the resulting interpretation of the rule meets their expectations of relevance. Subjects then select the cards that may test the consequences they have inferred from the rule. Order of accessibility of consequences and expectations of relevance vary with rule and context, and so, therefore, does subjects' performance. By devising appropriate rule-context pairs, we predict that correct performance can be elicited in any conceptual domain. We corroborate this prediction with four experiments. We argue that past results properly reanalyzed confirm our account. We discuss the relevance of the Selection Task to the study of reasoning.