ction and possible mitigation. In this project, we have built two computational models -- one to represent single pilot performance and one to represent crew performance. Both models have been built from prior cognitive task analyses coupled with empirical performance data. To make the task analyses tractable, we needed to focus on a particular aspect of flight. Given the large number of errors that occur during changes in vertical position, we chose to focus on the climb and descent phases. Cognitive task analyses were then developed using Natural-language GOMS (Goals, Operators, Methods, and Selection Rules), developed by Kieras (1997). NGOMSL is based on Card, Moran and Newell's (1983) conceptualization of task decomposition into a hierarchical set of goals, which can be operated on using a set of methods and selection rules. The technique allowed us to specify the alternative set(s) of actions that must be carried out in order to achieve a set goal (e.g., the activities that a pilo