In recent years there has been an increasing interest in the development of social robots for educational purposes. A presumed advantage is their physical presence in the learner’s referential world. However, it remains an open question whether this embodied presence of a robot can be utilized to support learning with robots’ gestures. The aim of our study was to program a social robot that tutors mathematics through verbal explanations supported by deictic, beat, and iconic gestures that align with explanations, and to analyse its effectiveness on learning outcomes. The learning outcomes entail cognitive outcomes and affective outcomes. A between-subject experiment was performed to research the effect of the robot’s use of supporting gestures on cognitive and affective learning outcomes compared to a robot that does not perform gestures or performs random gestures. In total, 78 children (Mage= 8.3 years) from Dutch primary schools participated. Results showed that the social robot effectively taught children mathematical equivalence by a significant increase in test scores between the pre- and post-test. No differences were found in test scores between participants in the supportive gesturing condition, the no gesturing condition, and the random gesturing condition. Participants showed high task and tutor engagement scores and head direction estimations revealed that participants were attentive towards the robot, but no differences were found between the three conditions. However, the gestures ‘helpfulness’ was rated significantly higher by participants in the supportive gesture condition than the random gesture condition, suggesting that children felt the gestures mattered.