Several computational models of emotions have been proposed to enable artificial agents to generate emotions of their own. However, there are barriers that limit the full capabilities of these models. One issue is the need to enable emotion generation in autonomous agents in wide range of interaction situations instead of designing specific scenarios. Additionally, it is not practically easy task to ‘effectively’ integrate other human characteristics in emotion generation process of artificial agents, which is essential for variation in behavioural responses of such agents. Moreover, although theoretically it is believed that appraisal variables are associated with emotion intensities, existing emotion literature does not offer a generalisable mechanism to computationally achieve such a mapping—thereby leading to ad-hoc implementations. It is also important to note that emotions expressed by intelligent autonomous agents like robots can have deep impact on people and society, therefore, it is crucial to ensure ethical implications of emotional responses of such systems. In this paper, we endeavour to review the emotion models proposed in the last two decades based on the aspects discussed above and provide recommendations for the development of future computational models of emotion. Our review will mainly revolve around the emotion models that implement the concept of appraisal theory of emotion. Our finding suggests that none of the existing computational models of emotion using appraisal theory implement all the characteristics we identify thereby providing further research opportunities.