Affective computing and emotional agents have been found to have a positive effect on humancomputer interactions. In order to develop an acceptable emotional agent for use in a self-service interaction, two stages of research were identified and carried out; the first to determine which facial expressions are present in such an interaction and the second to determine which emotional agent behaviours are perceived as appropriate during a problematic self-service shopping task. In the first stage, facial expressions associated with negative affect were found to occur during selfservice shopping interactions, indicating that facial expression detection is suitable for detecting negative affective states during self-service interactions. In the second stage, user perceptions of the emotional facial expressions displayed by an emotional agent during a problematic self-service interaction were gathered. Overall, the expression of disgust was found to be perceived as inappropriate while emotionally neutral behaviour was perceived as appropriate, however gender differences suggested that females perceived surprise as inappropriate. Results suggest that agents should change their behaviour and appearance based on user characteristics such as gender.