The current study reports a pre-registered investigation into the interrelations between mathematics anxiety, metacognition and mathematical decision-making. Although this question has already received some attention in previous work, reliance on self-report measures of metacognition has hindered its interpretation. Here, a novel experimental mathematical decision-making task was used in which participants solved mathematical assignments of varying difficulty, and expressed their level of confidence in the accuracy of their decision both prospectively and retrospectively. Mathematics anxiety was measured using a standardized questionnaire. Both prospective and retrospective confidence judgments predicted unique variation in accuracy; however, the explanatory effect of prospective confidence disappeared after taking task difficulty into account. This suggests that prospective, but not retrospective, confidence is largely based on easily available cues indicative of performance. Results of a multiple regression analysis indicated that individual differences in mathematics anxiety were negatively related to the overall level of confidence (both prospectively and retrospectively), and positively related to metacognitive efficiency (only prospectively). Having insight in these interrelationships is important in the context of remediating mathematics anxiety, which might in turn be useful with regard to the worldwide need for more workers with degrees in science, technology, engineering, or mathematics (STEM).