The purpose of this thesis is to develop a discrete event simulation model of post-earthquake restoration for the Los Angeles Department of Water and Power (LADWP) water supply system. Discrete event simulation, a new approach to modeling post-disaster lifeline restoration, offers many benefits for restoration modeling compared to alternative methods. The water supply system and restoration process are represented in great detail with few simplifications. The utility company's decision variables (e.g., number of repair crews, repair prioritization rules) are included explicitly, allowing exploration of their effects on the speed of the restoration. Restoration times are estimated separately for each region within the service area, and uncertainty in the process is modeled explicitly. With a service area of more than 1,200 km2 and 12,000 km of pipelines, the LADWP water supply system is the largest municipal system in the United States. Extensive review of the LADWP water organization, water supply system, and postearthquake restoration process was conducted. This review provided the basis for the restoration model. Crews, tasks, and the different phases in the restoration process came directly from discussions with LADWP personnel and the water organization's emergency response plans. For a particular earthquake, the restoration model takes as input information about damage to the system and the resulting hydraulic flow, both of which are provided by the Graphical Iterative Response Analysis for Flow Following Earthquakes (GIRAFFE) model that was developed for the LADWP system (Shi 2006, Wang 2006). Throughout the restoration simulation, the model interacts with GIRAFFE periodically in order to receive updates of the system functionality at specific times as the restoration process proceeds and damage is repaired. The restoration model provides several different types of output including system and subregion restoration curves; spatial distribution of restoration; material usage; crew usage; average time each customer is without water; and time to restore the system and subregions to 90%, 98%, and 100%. It can also include damage uncertainty by combining the output from runs for multiple realizations of damage associated with a single earthquake. The model can be used to help estimate economic and societal losses due to water supply system outages, and to evaluate the effectiveness of possible restoration improvement strategies. Ten simulations of the restoration model were run using real damage data from the 1994 Northridge earthquake as input, and the results were compared to the actual restoration that took place following Northridge. The average spatial distribution of restoration roughly matches what occurred in 1994. As in real life, the areas experiencing longer outages in the model are mainly in the north of the system service area or around the San Fernando Valley. The system restoration curves did not match exactly, as the range of outputs from all 10 runs of the restoration model shows that the restoration occurs too quickly, especially during the first day after the earthquake. Possible future model modifications that may improve the calibration are discussed. (Abstract).