Affective computing, the study of giving computers the ability to perceive human emotions and of endowing computers with synthetic emotions, has recently become an area of great activity. With the work of Mayer and Salovey, as well as that of Goleman, suggesting that emotions are to some extent quantifiable and can be studied with due rigor, work in the field of psychology has produced a large number of emotional models. Many computational architectures have been developed with the goal of synthesizing emotions for an artificial agent. These range from the affective dimensions in Breazeal's work to the goal based architecture developed by Michaud. The emotion-based architecture described in this thesis builds on previous work by Lisetti, combining scripts for emotion concepts based on semantic primitives and computational scripts based on the componential appraisal theory of emotions. The thesis describes some aspects of one implementation of the architecture on a specific platform as an emotional state generator for a robotic office assistant. The architecture itself, though, is not platform specific and could be implemented on different types of autonomous agents.