This study examines the use in Japanese of verb forms containing -(r)are in syntactical expressions. The meaning and function of the adversative passive and its behavior vis-a-vis the non-adversative or plain passive is discussed, and the related non-derived constructions and their relationships to the adversative passive are analyzed. Finally the implications for voice systems in general are considered. It is concluded that the morpheme -(r)are is not exclusively an inflectional marker, but works as a marker of canonical passivization in derived plain and affective passives much like the English passive--ed, and works as a limited derivational morpheme elsewhere. The morpheme appears in both 'basic' and 'derived' constructions and can fulfill either grammatical or semantical constructions or both at the same time, in the affective passive. Furthermore, semantic voice is not always realized inflectionally and pragmatic voice is not always realized by sentence derivation. To understand how voice works, one must look not just at morphology or sentence derivation, but at how the grammatical relations and semantic roles are affected in various constructions. (JL)