Requirements identification is a critical step in the development of any system. When the goal is to develop automation software to offload and augment tasks previously done by humans, the requirements identification challenge becomes one of understanding what experts do, why they do it, and when they do it. To accomplish this, a cognitive task analysis (CTA) was conducted to enumerate the what, when, and why of job performance for trainers at the Joint Warfighting Center (JWFC). The results are being used to help build intelligent agent software to aid trainers associated with Category 3 level training. A total of seventeen interviews were completed and the resultant data provided a wealth of information for compilation and review. This included crucial information that helped to define the intelligent agents software. For example, the CTA revealed types of instructional feedback the trainers give and when they give it. In addition to the knowledge used directly for software development, the results also gave insight regarding the practical considerations of using CTA. The purpose of this paper is to define the CTA problem space, discuss different approaches that were considered, and describe the methodology used for conducting the CTA. The paper will conclude with results and lessons learned that demonstrate the applicability of CTA to the project at hand, as well as to the larger community of training systems.