Over the past two decades, significant advances have been made in speech analysis and speech pattern recognition techniques, however, the penetration of these advances (notably in pattern recognition techniques) into the speech disorders research arena has lagged, and penetration into the clinic is virtually non-existent. Here we examine one approach to adapting and extending speech recognition technology based on Hidden Markov Modeling (HMM) to an analysis of speech from children with speech disorders of unknown origin. Specifically, we examine the use of normal-speech trained HMMs to identify acoustically defined categories of segmental distortions, and use those categories to characterize differences among a group of children with developmental speech delays.