Research on the educational applications of social robots has shown how they can motivate children and help improve academic learning outcomes. Here, we examine how robots can support skill learning and, more specifically, musical instrument practice. Drawing from social facilitation theory and evaluation apprehension theory we expected that the robot's mere presence would impact children's performance and that this effect would be contingent upon the children expecting the robot to evaluate their performance. We report an experiment with children (N = 31) aged nine to twelve who practiced a familiar and new piece alone, in the presence of an evaluative robot, and in the presence of a non-evaluative robot. We found that children performed better in terms of rhythm, pitch, and general impression in the presence of the non-evaluative robot. These findings offer important insights for designing robot tutors for music learning.