Two commonly applied methods of differentially classifying language-delayed children as either specifically language impaired or developmental lag language impaired (i.e., children with low IQ and low language performance) were examined for stability over 1-and 2-year periods. One classification method, following the DSM III.R guidelines, was based on an absolute cut-off level for performance on a measure of cognitive ability, in conjunction with other exclusionary criteria (i.e., language delay that is not the result of hearing loss, social– emotional disorder, etc.). The second classification method included the same absolute cut-off for cognitive ability as the DSM III-R guidelines, but it also required that a minimum relative difference be present between the general cognitive performance and the language performance. These two methods were examined for differences in classification of children, as well as for stability of classification across time. We used the McCarthy Scales of Children's Abilities Perceptual-Performance Index as the cognitive measure and the Peabody Picture Vocabulary Test-Revised and the Test of Early Language Development as language performance measures. Results indicated significant differences in classification between the two methods of defining SLI, as well as substantial changes in classification over time using either method. Alternative classification systems are discussed.