We consider two problems related to the understanding of facial expressions: (I) modeling the perception of human facial expressions from information extracted from face photographs and (II) reasoning about the transition between facial expressions as modeled in (I). For (I) we develop a hierarchical fuzzy modeling method. The results obtained are satisfactory when compared with those of other approaches: for the six basic facial expressions (happiness, surprise, sadness, anger, disgust and fear), we achieve a recognition ratio of 85 %. More importantly, in addition to the usual advantages of fuzzy rules such as easy interpretation and adaptability in a changing environment, our approach can be applied without additional training to (i) the identification of mixed facial expressions – the recognition ratio of the two strongest facial expressions in the same face being about 75%, and (ii) the recognition in noisy environment (that is, our approach can successfully deal with lack of data). For (II) we propose a model for transition between facial expressions based on a hybrid reasoning scheme combining generalized modus tollens and fuzzy backward reasoning. Appropriate examples are used to illustrate our approach and we conclude with a discussion about future issues, and possible applications to an intelligent humancomputer interface using the transition model.