If preference-inconsistent information initiates more effortful cognitive analysis than does preference-consistent information, then people should be more sensitive processors of information they do not want to believe than of information they do want to believe. Three studies supported this prediction. Study 1 found that inferences drawn from favorable interpersonal feedback revealed a correspondence bias, whereas inferences drawn from unfavorable feedback were sensitive to situational constraint. Study 2 showed this sensitivity to the quality of unfavorable feedback to disappear under cognitive load. Study 3 showed that evaluations of the accuracy of favorable medical diagnoses were insensitive to the probability of alternative explanation, whereas evaluations of unfavorable diagnoses were sensitive to probability information. The importance of adaptive considerations in theories of motivated reasoning is discussed.