This study investigates the underlying structure of the listening test of the Singapore-Cambridge General Certificate of Education (GCE) exam, comparing the fit of five cognitive diagnostic assessment models comprising the deterministic input noisy "and" gate (DINA), generalized DINA (G-DINA), deterministic input noisy "or" gate (DINO), higher-order DINA (HO-DINA), and the reduced reparameterized unified model (RRUM). Through model-comparisons, a nine-subskill RRUM model was found to possess the optimal fit. This study shows that students' listening test performance depends on an array of test-specific facets, such as the ability to eliminate distractors in multiple-choice questions alongside listening-specific subskills such as the ability to make inferences. The validated list of the listening subskills can be employed as a useful guideline to prepare students for the GCE listening test at schools.