This dissertation presents a new physics-based post-earthquake fire spread simulation model that estimates the extent of fire damage in an urban area, given a set of ignition locations and times. The key features of this model include (1) using actual building footprints and heights from remote sensing data and advanced GIS algorithms to generate specific rooms within a building so that room-to-room fire spread can be modeled, (2) calculating and using configuration factors from actual building footprints and spatial orientation of buildings to estimate the percent of radiation another building receives, (3) recognizing that a roof flame behaves different from a window flame and therefore, representing a roof flame as an open pool fire, (4) developing and using a brand generation model from empirical data and (5) implementing a new state-of-the-art ignition model which can be run deterministically or probabilistically based on a specific earthquake ground motion. A room-specific time-temperature curve describes the evolution of fire within each room. Room-to-room fire spread can occur through doorways to adjacent rooms, by burn-through to adjacent rooms or a room or roof above, or by leapfrogging through windows to a room above. Radiation flux from room gas, flames ejected out windows, and roof flames are all calculated to determine if fire spreads to neighboring buildings. Fire brand size and traveling distance are estimated to determine if fire spreads to neighboring buildings farther away. Each of these modes of spread is represented explicitly using models based on physical laws and empirical data, and including randomness in the process. Using a case study area in Los Angeles, results such as the percentage of area burned in a building, distribution of the spread modes, and a building?s time to fully burned are calculated. Sensitivity analyses are conducted for fourteen parameters, including the number and location of ignitions, wind direction and speed, building density, and average room length. The results of this new approach are compared with that of the widely used Hamada model.