This talk describes development of a physics-based mathematical and computational model to predict fire spread among structures and natural fuels (trees, shrubs and ground litter). This tool will be used to understand how fires spread in a community where both structures and natural fuels coexist, to help train fire fighters and to quantify the benefits of mitigation actions. No such model currently exists. There is an increasing awareness among fire fighters, community action groups and community planners of the need for such a model. This 'neighborhood-scale' model can use detailed data on the topography, local meteorology, building layouts and elevations, three-dimensional distributions of natural fuels, and the material properties of both the natural fuels and the structures. Nearly 10 % of the land and over one-third of the homes in the U.S. today belong to the Wildland/Urban interface (WUI), and these fractions are increasing rapidly. Fires in the WUI setting have also been increasing rapidly, becoming a national (as well as an international) problem. Models of the WUI fires must include the long-duration, high-intensity burning characteristics of structures as well as the burning characteristics of vegetation. Over the past 25 years, the Building and Fire Research Laboratory (BFRL) at the National Institute of Standards and Technology (NIST) has been developing a physics-based mathematical and computational model, known as the Fire Dynamics Simulator (FDS), to predict fire spread in a structure. This model is available free over the Web (www.fire.nist.gov), is well regarded and is widely used by fire protection engineers around the world. BFRL has recently extended the model to include fire spread from structure to structure and is now generalizing FDS to include prediction of fire spread in both continuous and discrete natural fuels. The current model, as well as its generalization, is both computationally and data intensive, requiring for any specified region, high-resolution, three-dimensional data of the quantities mentioned above. This talk will describe the physics and the computational methodology used in the model, the data and computational resources needed, and some results. Simulation of fire spread on a single plot of land (with one or two structures, trees, shrubs and combustible ground litter such as pine needles or leaves) will be shown, as well as fire spread in a small neighborhood, including several structures and wildland fuels. See the attached figure, which shows fire and smoke growth and spread computed in such a simulation. The model includes most of the mechanisms for fire spread at these length scales (fire spread by brands is not included). For these simulations to be predictive, fire spread from one fuel element (structure, tree or shrub) to another must be compared with data, although such data generally does not exist. Some implications on fire spread of the fact that structures have a much greater fuel load and a much longer ignition time than wildland fuels will be discussed. For example, entrainment of air by the plumes from multiple, fully involved, burning structures can substantially change the wind patterns and therefore the spread of the fire front at some distance from the structures.