Animated agents have the potential to increase engagement and learning during online training by acting as personalized tutors. However, little is known about the conditions that make these agents most effective. In this study, 183 e-learners completed a Microsoft Excel training course. Approximately half were assigned an agent with predetermined features. The others were allowed to choose their agent's appearance, personality, feedback style or all of the above features. Offering multiple choices increased learning. Unexpectedly, choice of feedback style alone decreased self-efficacy. Choosing the agent's appearance increased self-efficacy and the number of training modules completed. Overall, this study expands the learner control literature, identifying a new form of learner control that has some beneficial effects on knowledge acquisition. (Contains 4 tables and 4 figures.)