In this paper, we meta-analyse the empirical evidence about the effectiveness of game-based interventions to reduce students' level of math anxiety. After performing a search for randomised controlled studies relevant to game-based intervention for math anxiety, 16 experimental studies with a total of 686 participants described in 11 peer-review articles met the selection criteria.A random-effects meta-analysis indicated a small and non-significant reduction of math anxiety (mean effect size ES=-0.32, CI=[-0.64,0.01]). The results were moderated by several factors: non-digital games were more effective, while digital games had a negligible mean effect size of $ES=-0.13$, $CI=[-0.33,0.08]$. The effect size was moderated also by the total duration of the intervention, to the advantage of longer interventions, and by the type of gameplay: games had a greater effect on math anxiety reduction when they promoted collaborative and social interactions. Such features were only present in non-digital games, while all the digital games analysed were single-player. In the final section of the paper, we discuss future possible research directions. The weak results obtained indicated the need to develop and test games explicitly designed for math anxious students. This will require the investigation of the relationship between game features and math anxiety through the analysis of the behaviour of anxious and non-anxious students at play. Among the features that an anxiety-aware game could employ, we suggest collaborative gameplay, social interactions, adaptability, features promoting intrinsic motivation and real-time measurements of math anxiety.