The present study investigates higher-order factor structure among fifteen primary variables selected from four broad noncognitive domains of academic self-beliefs, motivation, learning strategy, and attitudes toward school. The PISA 2003 international dataset was analyzed. Several EFA, CFA, and SEM models were tested, hypothesizing the structure among the primary first-order variables and their relationships to the mathematics scores. The analyses indicated no single, general factor at the second-order level, encompassing all fifteen first-order variables. Instead, the fifteen primary variables were best represented by a three-level factor structure with the four salient domain factors at the second-order level and one general noncognitive factor at the third-order. The most plausible SEM model had each of the three self-belief primary variables individually linked to the mathematics achievement scores, independent of the third-order factor. Self-efficacy was the strongest predictor of mathematics achievement and its predictive power was comparable to that of the common part of all 15 primary variables captured by the general noncognitive factor.