This paper addresses auditors' hypothesis testing during diagnostic inference. Cognitive steps involved in diagnostic inference include mentalrepresentation formation, hypothesis generation, and hypothesis testing (Einhorn [1976]). Hypothesis testing can be decomposed further into specifying questions (or tests), interpreting answers (or feedback), and revising hypotheses (Libby [1985]). Within two main experiments we study the soundness of, and how incentives influence, auditors' question specification. Four types of unsound reasoning are considered: confirmation and disconfirmation proneness (i.e., unwarranted hypothesis retention and abandonment), information proneness (i.e., significant valuation of evidence independent of its expected diagnosticity), and affirmation proneness (i.e., overvaluations of "yes" versus "no" answers, holding constant their expected diagnosticity).