Two experiments designed to examine the specificity of emotional source memory are reported. In the encoding phase, participants saw faces along with emotional context information, that is, descriptions of cheating, trustworthy, or irrelevant behavior. In the test phase, participants were required to complete a source classification test and a cued recall test. In both experiments, the source memory advantage for faces characterized by negative context information (cheating) was replicated. Extending previous research, a multinomial source-monitoring model was applied to distinguish between specific source memory for individual behavior descriptions and partial source memory in the sense of only a rough classification of the behavior as belonging to a particular emotional category—cheating, trustworthy, or neither of these. The results indicate that the source memory advantage for the emotional context information is not always accompanied by enhanced recollection of the specific details of the learning episode and might rather reflect unspecific memory for categorical emotional information.