Two models of hierarchical decision making in a conflict detection task were compared (Niessen, Leuchter, & Eyferth, 1998; Rantanen & Nunes, 2005) using eye-tracking data. 30 experienced air traffic controllers individually processed a series of 56 trials. In each trial, 15 participants had to decide whether two aircraft displayed on a simulated radar screen would be in conflict, i.e., whether their horizontal and vertical distances would eventually violate legal separation rules, the 15 others had to decide whether an intervention would be required to avoid a conflict. Results showed that none of the two models could fully account for the data. These results led us to produce a new model of hierarchical decision-making, implemented in the cognitive architecture ACT-R 6.0.