a b s t r a c tAffective responses of users to system messages in human-computer interaction are a key to study user satisfaction. However, little is known about the particular affective patterns elicited by various types of system messages. In this experimental study we examined if and how different system messages, presented in different modalities, influence users' affective responses. Three types of messages, input requests, status notifications, and error messages, were presented either as text or speech, and either alone or in combination with icons or sounds, while users worked on several typical computer tasks. Affective responses following system messages were assessed employing a multi-modal approach, using subjective rating scales as well as physiological measures. Results show that affective responses vary systematically depending on the type of message, and that spoken messages generally elicit more positive affect than written messages. Implications on how to enhance user satisfaction by appropriate message design are discussed.