A two-stage hierarchical classification scheme of psoriasis lesion images is proposed. These images are basically composed of three classes: normal skin, lesion and background. The scheme combines conventional tools to separate the skin from the background in the first stage, and the lesion from the normal skin in the second stage. These tools are the Expectation-Maximization Algorithm, the quadratic discrimination function and a classification window of optimal size. Extrapolation of classification parameters of a given image to other images of the set is evaluated by means of Cohen's Kappa coefficient.