This paper examines the validity of diagnostic categories frequently used to classify children with severe language disorders by determining the relationship of the categories to independently derived developmental, psychological, and medical variables. It is argued that the classification systems currently available too often fail to achieve the goals of description and usefulness in promoting effective treatment. Case histories and other data from 82 children with severe language disorders were used to assign each child to one of seven diagnostic categories. The 82 cases were then regrouped on the basis of 32 clinical variables using a step-wise clustering procedure. The seven original categories collapsed to only two definable clusters. A repetition of the analysis using a larger sample and different variable set resulted in similar clusters. The variables which define these clusters are discussed, as well as the resulting implications for assessment procedures in general. (Author/AM)