The operation Agree is generally considered to lack direct semantic effects. Based on the Japanese object-honorific construction, this work claims that Agree can in fact interact with LF interpretation such that some semantic predicates select their arguments via Agree. More specifically, I claim that the object-honorific marker serves as a semantic predicate honor, which selects its arguments via Agree.Based on the distribution of the NP that can trigger object-honorific marking, the object-honorific marker seems to access the object via Agree, as previous studies have argued; I examine, however, the theoretically overlooked observation that object honorification is felicitous only when the object referent is honored by the subject referent as well as the speaker. Based on this observation and the result from previous studies, I argue that object-honorific morphology is semantically a predicate that takes two arguments, an honorer and an honoree, and finds those arguments via Agree.