This study conducted an item-level analysis of mathematics anxiety and examined the dimensionality of mathematics anxiety in a sample of developmental mathematics students (N = 162) by Multi-dimensional Random Coefficients Multinominal Logit Model (MRCMLM). The results indicate a moderately correlated factor structure of mathematics anxiety (r = 0.679). Developmental mathematics students were more anxious about tests/exams than other activities presented in the Mathematics Anxiety Questionnaire (Wigfield & Meece, 1998). Regression analysis further suggests that lowering mathematics anxiety level by 1 logits could raise developmental students' mathematics performance up to 8.768%. Mathematics anxiety explained 15.35% variation of mathematics performance (adjusted R[superscript 2]: 0.1535). Interventional strategies and implications for educational practices are discussed.