This paper describes a pattern classification model called “classification on subclasses”. The model and its computation scheme are based on the theoretic foundation of minimizing the cross-entropy of the distribution functions that bear considerable complexity and non-linearity. In this model, pattern classes are configured and described by a number of subclasses each associated with a distribution formulated according to the regularization principle. This modeling technique provides a simplified solution to a group of non-linear pattern classification problems. Simulation shows a high classification rate on pattern samples with complex distributions.