The combinatory possibies in V-V lexical compounds in Japanese are examined. The previous account of Kageyama (1993), who proposed that members of V-V lexical compounds must match in argument structure type, is shown to embody two problems. One concerns his criteria for distinguishing between unergative and unaccusative verbs, and the other is the existence of counterexamples to his claim (e.g., yomi-tsukareru (read-be.tired) ‘be tired from reading’). An alternative account of the constraints on possible lexical compounding is proposed. It is shown that the combinatory possibilities of lexical compounds are constrained by 1) the principle of subject sharing (Yumoto 1996) and 2) the conditions on the semantic structures of verbs (Matsumoto 1996), both of which are stated at the level of the semantic structure.