The monitoring of responses and reinforcement of correct responses are time consuming aspects of traditional speech therapy. A microcomputer-based system has been developed which will relieve the therapist of much of the burden in repetitive practice sessions. It has been designed around commercially available systems which permit the recognition of isolated words or short phrases and utilizes a computer-aided instruction format in the patient practice sessions. A discriminant analysis approach is used for word recognition rather than the more common template comparison technique. Discriminant functions are formulated for each word to be used in the practice sessions based on the productions (classified satisfactory or unsatisfactory) of the target words elicited by a therapist in initial sessions. A pilot study has shown that this approach permits common phonemic developmental articulation errors, such as omission of an initial /s/, to be detected with very high reliability. Discriminant functions prove to be more efficient in both memory and speed than template matchings for the practice sessions and this is a major consideration in a cost effective system. This system should prove to be inexpensive enough to permit widely decentralized installation and is currently being considered for use in speech therapy services of a major city school system.