This paper presents an analysis of the effects of the timing of emotional expression in human-robot interaction, exploring the hypothesis that emotional expression in human conversation is largely a matter of social practice and hence conventionally required at specific places in dialog. To identify when and how humans express emotions while speaking, a pre-study was carried out in which participants produced the robot’s utterances in dialog with another human interaction partner. Two coders analyzed participants’ non-verbal behaviors in these dialogs and translated these into non-verbal behaviors of the robot. A between-subject experiment was then carried out in which participants evaluated a social robot that provided emotional expression either after its utterances or based on the timing of emotional expression by human speakers. Participants evaluated the robot’s competence, warmth and their discomfort in the interaction, and responded to two behavioral variables that provide evidence of the robot’s persuasiveness. The results show that participants consistently produce emotional expression at similar places in dialog, confirming the conventional nature of emotional expression, but reveal only subtle effects on human-robot interaction.