By adopting a pre- and post-test design, the current study longitudinally examined the complex relationship between two different dimensions of phonological vocabulary knowledge (declarative vs. automatized) and their ultimate impacts on global L2 listening proficiency among 133 Japanese EFL students. The declarative group focused solely on what target words sound like and mean via meaning recognition tasks. The automatization group worked not only on such form-meaning mappings but also on prompt access to the target words in a semantically, collocationally, and grammatically appropriate manner via lexicosemantic judgment tasks. Compared to the declarative group, the automatization group showed relatively robust learning in both declarative and automatized dimensions of target words. Although neither training approach showed clear superiority, the results suggest that relative gains in automatized, rather than declarative, dimensions are associated with enhanced L2 listening proficiency. The distinction between declarative and automatized dimensions of phonological vocabulary knowledge, along with the absence of a direct link between training type and improved listening proficiency, offers valuable insights for future extension studies.