In recent years more and more emphasis has been placed on model discrimination procedures. In this paper we propose some new procedures for the selection of the most adequate regression model. Properties of those procedures are analyzed and compared. Their relationship with classical informal procedures is fully discussed. Our procedures are called the information criteria because we base our loss function on the Kullback-Leibler information measure of the distance between two probability density functions. The basic framework of our approach was originated by Akaike in his sequence of papers.