A discussion of computer-based language instruction examines the quality of materials currently in use and looks at developments in the field. It is found that language courseware is generally weak in the areas of error analysis and feedback, communicative realism, and convenience of lesson authoring. A review of research under way to improve lesson quality suggests preferred solutions in each of these areas. In the case of error analysis, where feedback is important to effective and efficient learning, the best hope for systematic, grammar-sensitive error analysis seems to be based on parsing techniques borrowed from natural language processing. Communicative realism can be enhanced by both simulation designs and interactive video, even though materials using the approaches are still rather crudely designed. Finally, a promising approach to convenient lesson authoring involves adaptation of word processing techniques to create analogies with textbook authoring. A brief list of references is included. (Author/MSE)