SimStudent is a computational model of learning with its cognitive fidelity of learning being demonstrated especially in the way it makes human-like errors. Using SimStudent as a teachable agent in an interactive peer-learning environment, we have investigated how tutee (i.e., SimStudent) learning af- fected tutor (i.e., human student) learning. In this paper, we are particularly interested in how tutees’ shallow learning af- fects tutor learning. We are also interested in how the errors that the tutee makes affect tutor learning. The results show that teaching SimStudent on a fixed set of problems makes students easy to tutor SimStudent, which in turn helps stu- dents learn, but is likely to allow SimStudent to commit shal- low learning, which is harmful for tutor learning. It is thus crucial to let the student detect SimStudent’s shallow learning and extend teaching until SimStudent and the student achieve satisfactory competence.