This study examined the influence of affect on conversation behavior and moderated effect of partner's affective state on the influence. Previous studies have focused on how conversation behavior is influenced by one's affective state. This study examined whether the relationships between one's affective state and ranking message (Study 1) or free description (Study 2) would change through the partner's affective state, in addition to previous studies that considered only the main effect of one's affective state. Through the two studies, vignette experiments were conducted and partner's affective state was represented using facial expressions (smile or frowning face) in abstract paintings. In Study 1, 158 undergraduates prioritized four categories of message (edification, disclosure, question, and acknowledgement). These were extracted from Verbal Response Modes (Stiles, 1992) for five contexts composed of five categories including four that were the same as above and another, “non-message”. In Study 2, 42 undergraduates provided free descriptions for the same five contexts. As a result, although there were few significant differences of correlation coefficients between facial expressions in study 1, there were some significant differences in Study 2. These results demonstrated the benefit of considering influences of dyadic combination of affective state on conversation behavior.