The Formosan language Tsou embraces three serial verb constructions where verbs bearing nonagreeing voice marking are strung together. While the combination of a nonactor voice first verb and an actor voice second verb appears to be an exception to the widespread tendency that components within a serial stricture have the same value for grammatical categories typically associated with verbs, this study argues that the nonconcordant voice marking actually reflects the syntacticosemantic conditions for integrating verbs as well as the event descriptions in their denotations into a joint predication, and adjustment in the pattern of argument realization jointly attests to the presence of these conditions. In Tsou, a nonconcordant verb series that introduces thematically peripheral, nonsubcategorized entities into a monoclausal structure is only well formed if the participating verbs do not clash in event properties crucial to the Vendler-Dowty style of lexical aspectual classes. On the one hand, constraints on the use of voice marking and combinatorial possibilities of verbs show that serialized verbs share common specifications for telicity and duration. On the other, changes in the morphosyntactic expression of arguments illustrate the possibility to shift aspectual characteristics in the verb's denotation so as to bring event descriptions within the entire verb series to a unified integration. As a whole, the Tsou data make evident the presence of event integration within verb serialization, even though the nonconcordant voice marking appears to suggest otherwise.