This study explored whether a virtual agent in a feedback task would enhance social presence and motivation, as well as facilitate learning on a Chinese symbol retention task. Learning occurred with four different computer-based feedback sources: a computer-generated “Agent”, a “Human”, an “Inanimate” image and a non-image “Verbal” condition. Measures of presence, motivation, and learning were generally related, albeit with limited support for an Agent compared to other sources of feedback. Although performance with the Agent was comparable to the Human on several measures, it did not consistently differ from other conditions. Agent and Human conditions did elicit highest social presence, but this was not uniquely related to learning outcomes. Surprisingly, the Verbal condition resulted in optimal learning. Results suggest that the utility of virtual agents is dependent on task relevance and on subjects’ inference about the agent’s capacity; agent characteristics can be distracting as well as facilitating.