Five experiments were conducted to address the question of whether source information could be accessed in the absence of being able to recall an item. The authors used a paired-associate learning paradigm in which cue-target word pairs were studied, and target recall was requested in the presence of the cue. When target recall failed, participants were asked to make a source judgment of whether a man or woman spoke the unrecalled item. In 3 of the 5 experiments, source accuracy was at or very close to chance. By contrast, if cue-target pairs were studied multiple times or participants knew in advance of learning that a predictive judgment would be required, then predictive source accuracy was well above chance. These data are suggestive that context information may not play a very large role in metacognitive judgments such as feeling-of-knowing ratings or putting one into a tip-of-the-tongue state without strong and specific encoding procedures. These same results also highlight the important role that item memory plays in retrieving information about the context in which an item was experienced.