Current evidence suggests that math anxiety and working memory govern math performance. However, these conclusions are largely based on simple correlations, without considering these variables as a network or examining correlations at the latent variables level. Thus, questions remain regarding the role of the unique and shared variance between math anxiety, working memory and math performance. The purpose of the current study was to (i) uncover the underlying relationships between the variables to understand the unique contribution of each element to the network; (ii) measure the shared variance and identify the interactions between affect and cognition that control math performance. Our analytical approach involved both network analysis approach and structural equation modeling, with a sample of 116 female students.Results show that math anxiety and working memory affect math performance by different mechanisms. Only working memory tests that included numeric information were correlated to math anxiety. Each of the various working memory tasks correlated differently to math performance: working memory as a single latent variable was a better predictor of math performance than visuospatial and verbal working memory as two separate latent variables. Overall, both working memory and math anxiety affect math performance. Working memory tasks that include numeric information can affect performance in math anxious individuals.