The paper begins by reviewing some problematic issues in the evaluation of educational software. Special attention is drawn to the formal and practical difficulties associated with "classical" input/output evaluation paradigms. These problems are especially acute in the case of the vast range of computer-based materials which do not readily fall into the conventional CAL mould. Such software includes adventure games, word processors, spreadsheets etc. We examine the evaluation process in terms of its clientele. We suggest that administrators, working at the strategic level of decision making, and teachers working at the tactical level, have needs for rather different types of information, which cannot necessarily be met through the application of any single approach to evaluation. We identify our particular clientele as being classroom teachers, and we discuss the implications of this targeting for the design of evaluation procedures. We discuss the central role of the practitioners themselves in generating an appropriate richness and variety of information. We describe the development of "IN-CITES": a classroom-based case study approach to software evaluation intended for application by autonomous self-programming groups of teachers. This approach is based on the so-called "triangulation" methodology, and employs both methodological triangulation (using a range of techniques) and observational triangulation (using a variety of observational perspectives). Our approach is designed to provide teachers with a framework for software evaluation, rather than a rigidly denned set of procedures. We acknowledge that many teachers have specialised knowledge and skills which they will wish to bring to bear on their own problems, and we have worked to provide the outlines of a disciplined approach to the evaluation process analogous to the case history method in routine medical practice.