This paper describes some results of a collaborative effort between the University of Pittsburgh and the Air Force to develop advanced troubleshooting training for F-15 maintenance technicians. The focus is on the cognitive task methodology used in the development of three intelligent tutoring systems to inform their instructional content and approach, and how task analysis results are reflected in particular features of one of these tutors. A well-conducted cognitive task analysis (CTA) can give a system developer information about the knowledge and skills students find difficult. In the three developed systems, a CTA methodology called PARI (Precursor (goal), Action, Result, and Interpretation) (3) informed the design of coaching and postproblem reflection. These systems, Sherlock, Hydrive, and Eaglekeeper, are designed to give those who maintain F-15 aircraft feedback about their reasoning errors and violations of good troubleshooting practice. Examples show how the PARI methodology informed the development of these systems, and results with 18 tutored novices, 23 untutored novices, and 13 master technicians are presented to show the efficacy of the training. The two fully developed systems are proving efficient and practical in improving student performance. The third, Eaglekeeper, remains in an earlier stage of development. Work with these systems illustrates the importance of early CTA to save time and effort in system development. (Contains 10 figures.) (SLD)