Mass fire-spread is a potential threat to some densely built urban areas by its devastating destructions. Especially in case of a large earthquake when multiple fires break out simultaneously, fire-spread hazard is likely to overwhelm fire-fighting capabilities and enlarge damage area. To explore fire-spread behavior and assess its damages, a simulation model of fire-spread in densely built urban areas is developed. Cellular Automata (CA) is a Bottom-up dynamics model that can reproduce a complicated phenomenon by setting up simple rules in a cell space. However, the traditional grid-based cells of CA are not suitable for modeling building-to-building fire-spread behavior. Therefore, an irregular coarse CA schema is proposed in this paper. Two sub-processes are involved urban mass fire-spread, i.e. (I) fire-developing in a single building and (II) fire-spread among buildings. When a fire is developing in a single building, the building will experience 5 fire stages along time, which become the states of cell. While fire spreads among buildings, there are 2 spread patterns: (I) short-range direct flame contact, radiation and convection spread and (II) long-range firebrand spotting spread. In relation to the 2 spread patterns, 2 sets of neighborhoods and rules of CA are formulated respectively. To verify the newly developed model, 100 times of individual random simulations for a real site fire-spread in Kobe City after Hanshin Earthquake (1995, Japan) are performed using an integrated GIS-CA-fire tool developed in the paper. Comparing the simulation results with the local observations, the general feature of fire-spread is found similar, and it is proved that the newly developed model is reliable to simulate urban fire-spread. Furthermore, based on the simulated results, a loss assessment model is formulated to calculate economic and life losses after fire-spread.