The testing and teaching of listening has been partially guided by the notion of subskills, or a set of listening abilities that are needed for achieving successful comprehension and utilization of the information from listening texts. Although this notion came about mainly through applications of theoretical perspectives from psychology and communication studies, the actual divisibility of the subskills has rarely been examined. This article reports an attempt to do so by using data from the answers of 916 test takers of a retired version of the Michigan English Language Assessment Battery listening test. First, an iterative content analysis of items was carried out, identifying five key subskills. Next, the discriminability of subskills was examined through confirmatory factor analysis (CFA). Five independent measurement models representing the subskills were evaluated. The overall CFA model comprising the measurement models showed excessively high correlations among factors. Further tests through CFA resolved the inadmissible correlations, though the high correlations persisted. Finally, we made 23 aggregate-level items which were used in a higher-order model, which induced best fit indices and resolved the inadmissible estimates. The results show that the subskills in the test were empirically divisible, lending support to scholarly attempts in discussing components in the listening construct for the purpose of teaching and assessment.