A broad empirical literature demonstrates what has been termed a confirmation bias or positive test strategy heuristic in reasoning (Klayman & Ha, 1987), a potentially maladaptive pattern of data preferences that coexists with more normative preferences for highly diagnostic information (Skov & Sherman, 1986). A model is developed to account for these variations in test strategies, beginning with the premise that cognitive processes are adapted to reducing particularly costly errors rather than to detecting "truth" (Cosmides & Tooby, 1992). By specifying the information required to minimize various errors of primary concern, the model clarifies the adaptiveness of certain confirmatory preferences, identifies conditions under which such preferences should diminish, and outlines how error minimization goals might produce data preferences coincidentally consistent with normative prescriptions.