This paper examines the problem of facial expression analysis in an evaluation-adaptation framework. In this paper, we are proposing a new approach, where the basic method is combined with a (post-processing) evaluation subsystem, which evaluates the obtained features and defines facial features that are to be refined. Anatomical constraints related to both 2-D and 3-D facial models are used for this purpose. Experimental results are presented, obtained in the framework of the ERMIS IST project, which illustrate the suc-cess of the proposed approach for extraction of facial features and expression analysis from low quality real life data.