As Langacker (1987, 2008) and Croft (2001) claim, there is no clear-cut boundaries exist between lexicon and syntax. In this vein, the present study aims at showing that Japanese lexical compound verbs [V1 V2]V (see Kageyama 1993), can be best captured as “constructions” within the framework of Construction Morphology (Booij 2010). Japanese lexical compound verbs possess non-compositional properties, which can be found in both morphology and semantics. With regard to morphology, there are “bound morphemes,” which appear only as part of a larger word. As to semantics, compounds possess “bound meaning,” the new and quite specific meanings of V1 or V2 when embedded in compounds (Booij 2010: 61). Moreover, there are a number of non-compositional compound verbs whose meaning cannot be synthesized from the meaning of their components. Examining these non-compositional compound verbs in a usage-based approach, this study shows that non-compositional compounds require high token frequency to support their meaning to be maintained and shared by a language community. Most importantly, the rule-based (top-down) approach in previous studies cannot explain these non-compositional properties. Instead, the usage-based (bottom-up) approach such as Construction Morphology can provide an adequate account for both compositional and non-compositional properties of compound verbs.