It is well-established that a negative relation exists between math anxiety (MA) and math performance. However, because there has been no systematic quantitative research on the predictors of MA, we conducted three studies in adolescents and adults to address this gap. Focusing on the Polish population, we tested whether, and to what degree, domain-specific (mathematical resilience, intellectual helplessness in mathematics, math performance) and domain-general (sociodemographic: gender, age; affective: general anxiety, test anxiety; and cognitive: fluid intelligence, working memory, response inhibition) variables predict and explain variance in MA. We found that regardless of the sample and other variables included in the models, intellectual helplessness in mathematics and mathematical resilience are consistent and independent predictors of MA. Moreover, math performance, rather than math grades, serves as a consistent predictor of MA. The findings highlight the relative importance of these variables in understanding MA and indicate a need to focus on domain-specific variables in targeting MA reduction.