Pupils’ poor achievement in mathematics has recently been a concern in many countries. To address this issue, it has been proposed to teach chess in schools. However, no convincing evidence of the benefits of chess instruction has ever been provided, because no study has ever controlled for placebo effects. This study implemented a three-group design to control for placebo effects. Measures of mathematical and metacognitive skills were taken. The results showed that the chess-treated group achieved better scores in mathematics than the placebo group (attending a Go course) but not than the control group (attending regular school lessons). Regarding metacognition, no differences occurred between the groups. These results suggest that some chess-related skills generalize to mathematics, because chess compensated for the hours of school lessons lost, whereas Go did not. However, this transfer is not mediated by metacognitive skills, and appears to be too limited to offer educational advantages.