Abstract: Airlines constantly face operational problems that develop from equipment failures, labor inabilities, or adverse weather conditions. These problems cause airline schedule disruptions, such as flight delays, flight cancellations and passenger mis-connections. When encountering such disruptions, airlines adapt various strategies, which form an airline schedule recovery process, to mitigate the impacts of these problems. An airline schedule recovery process is a sequence of actions to reallocate and reassign resources in order to minimize total impact (e.g. passenger delays, operational costs) after one or more incidents have disrupted the operations. The airline schedule recovery process is a collaborative effort from various organizations within an airline. Airlines typically have two main decision-making organizations: airline operations centers (AOCs) and station operation centers (SOCs). These two organizations are responsible for the flight schedule control; they work together to ensure schedule punctuality and to minimize the impacts of disturbances. The AOC, located at the airline headquarter, is in charge of en-route operations control, aircraft re-routing, and system-wide optimization evaluation. The SOCs, located at major hub stations, are responsible for gate and ground resource assignments, ground holds, and re-assigning connecting passengers. This research analyzes the delay recovery decision-making process that aims to minimize the total passenger delay. We compare two approaches, to the airline schedule recovery problem: a centralized approach and a bi-level approach. Each is designed for a different a organizational decision-making structure. The bi-level approach splits responsibilities between the AOCs and SOCs, whereas in the centralized approach, the AOC is the only decision-maker. Deterministic mathematical optimization models are used to determine the best strategic moves, such as canceling flights, delaying flights and ferrying reserve aircraft. The models are applied to disruptions of varying duration and time of day of occurrence. The solution for each model specifies the modified flight schedules as well as the (re)routing of non-stop and connecting passengers. A case study of a large U.S. carrier is used to understand the differences between the two approaches, and to evaluate their computational efficiency. The results indicate that the central model of decision making always outperforms the bi-level model. This is true in terms of total passenger delay and schedule recovery. Although both decision making structures result in similar flights delay and cancellation actions, the centralized model tends to delay flights more strategically in such a as to retain more of the schedule banks at the hub.