In the past decades, human computer interaction field began to witness studies exploring not only the rational characteristics of human users in making decisions but also their “extra-rational” aspects in interacting with environment and devices. The first task in exploiting one such facet, human affective states, is to accurately recognize and assess the internal state affective state. In this paper, we summarize the features, challenges and requirements of real-world HCI systems to user modeling and assistance. A survey of the state-of-the-art of affective state assessment focuses on computational assessment models underlying different studies. A user affective state assessment framework developed by the author based on dynamic Bayesian networks is also visited. The last part of the paper moves on to articulate what is perceived the future research direction contributing to the challenging research of affective state assessment in user modeling, which foresee integration of multi-disciplinary knowledge bases.