Previous studies have suggested that discrepant reporting in a test-retest reliability paradigm is not purely random measurement error, but partly a function of a systematic tendency to say "no" during retest to questions answered positively at initial testing ("attenuation"). To examine features of interview questions that may be associated with attenuation, three raters independently assessed the structural and content features of questions from the Diagnostic Interview Schedule for Children (version 2.3) and linked these to data from a test-retest reliability study of 223 community respondents (parent and child reports). Results indicated that for both parent and youth reports, item features most strongly associated with attenuation were (a) being a "stem" question (asked of all respondents, regardless of any skip structure); (b) question placement in the first half of the interview; (c) question length; (d) question complexity; or (e) requiring assessment of the timing, duration, or frequency of a symptom. Findings may be explained by participants' conscious efforts to avoid further questions or by their learning more about the nature and purpose of the interview as they gain more experience; alternatively, findings may represent a methodological artifact of structured interview design.