Pointing to declines in self-reported criminality across waves of the National Youth Survey, several researchers have concluded that "testing effects" may render longitudinal self-report data unreliable. This article argues that the issue remains unsettled on two accounts. First, alternative explanations for the declines have not been fully addressed. These include matters of scale construction, item-specific age-crime curves, and selective attrition. Second, previous research tends to conflate two types of testing effects explanations: panel fatigue and changing content validity. Each of these five explanations has different implications and is explored in the present article. Through this series of analyses, the author concludes that the declines stem from item-specific issues, namely, the inclusion of early-peaking offenses in the scales and the changing content validity of some survey items. Implications are discussed with respect to how criminologists operationalize key constructs such as crime and deviance and how we study the age-crime relationship. (Contains 2 tables, 6 figures and 7 notes.)