In this article, we explore how people revise their belief in a hypothesis and the reliability of sources in circumstances where those sources are either independent or are partially dependent because of their shared, common background. Specifically, we examine people's revision of perceived source reliability by comparison with a formal model of reliability revision proposed by Bovens and Hartmann (2003). This model predicts a U-shaped trajectory for revision in certain circumstances: If a source provides a positive report for an unlikely hypothesis, perceived source reliability should decrease; as additional positive reports emerge, however, estimates of reliability should increase. Participants' updates in our experiment show this U-shaped pattern. Furthermore, participants' responses also respect a second feature of the model, namely that perceived reliability should once again decrease when it becomes known that the sources are partially dependent. Participants revise appropriately both when a specific shared reliability is observed (e.g., sources went to the same, low quality school) and when integrating the possibility of shared reliability. These findings shed light on how people gauge source reliability and integrate reports when multiple sources weigh in on an issue as seen in public debates.