In this chapter we will discuss the process of evaluating the Intelligent Tutor, the needs analysis for and the development of which was described in Chapter 5 of this book. In literature on CALL (Levy, 1997b; Goodfellow, 1999) or SLA research (McDonough & McDonough, 1997) we often find references to two types of evaluation: formative and summative. While the former can be said to accompany development projects, the latter is meant to follow them. This chapter will reflect on both concepts while trying to apply them to the Intelligent Tutor. It will also explore the research aspects of this evaluation. Having the software evaluated by a sample of student population other than the one it was designed for means that a lot of data will be presented about that population, which has no other purpose but to make sure that there is a good overlap between the two. Interesting as it is, this data will not be extensively commented on here, but may be explored in its own right within a separate publication, while the focus of this chapter remains firmly on the evaluation of the Intelligent Tutor. The findings indicate that there is a good match between the two populations and that the software itself is effective in terms of learning outcomes. In Chapter 2 of this volume we discussed the potential for research in CALL development projects. It would appear that the evaluation phase in addition to the needs analysis phase provides ample opportunities for research (Fischer, 1999; McDonough & McDonough, 1997). However, the primary goal of CALL software evaluation is not necessarily research, but the quality assurance function (Thayer & Dorfman, 2000) as an integral part of the software development cycle (Levy, 1999; Sommerville, 2001; Pfleeger, 1998). Thus CALL software is subject not only to pedagogical standards, but also to those of software engineering. This is the reason why, as pointed out by Chapelle (2001: 51), CALL is scrutinised ‘beyond what is expected of evaluation of other classroom activities’. While saying that the tendency to view CALL as some sort of experiment, as Chapelle (2001: 51) did within the context of the above quotation, is a fair assessment, this reason alone does not justify the amount of scrutiny it is exposed to. It is rather a requirement imposed on it by its dual nature:(1) computational, by virtue of which it has to satisfy the standards