Social science research relies heavily on self-reported data. However, it is known that self-reports, especially of sensitive behaviors, tend to be biased. Among many endeavors to address self-reporting bias, using informants (such as peers, co-workers, and family members) to provide alternative reports has become an increasingly popular approach that offers some special benefits. In this paper I argue that studying informant accuracy both helps deepen our understanding about how perceptions of others are configured and also helps develop better methods to use informant reports to correct self-reporting bias. I propose a general framework that links informant accuracy to not only reporter’s characteristics, but also alter’s characteristics, dyadic characteristics, and features of the object being reported on. To illustrate the framework, I apply it to analyzing self-reports and peer-reports of smoking among 4,094 middle school students in China. The analyses present novel strategies to validate behavior when the truth is unknown, confirm previous findings and illustrate selected mechanisms, discover new findings, and reveal the distinctive logics for identifying the presence and the absence of a behavior. The results also indicate that weighting informant reports by informants’ network centrality can more effectively correct self-reports. Lastly, I discuss the practical implications of this study, from health monitoring to online rating, and a Bayesian model to model self-reports, peer-reports, and the latent behavior simultaneously.