A new approach to modeling regional air chemistry is presented for application to industrialized regions such as the continental United States. Rural chemistry and transport are simulated using a coarse grid, while chemistry and transport in urban and power plant plumes are represented by detailed subgrid models. Emissions from urban and power plant sources are processed in generalized plumes where chemistry and dilution proceed for 8--12 hours before mixing with air in a large resolution element. A realistic fraction of pollutants reacts under high-NOx conditions, and NOx is removed significantly before dispersal. Results from this model are compared with results from grid models that do not distinguish plumes and with observational data defining regional ozone distributions. Grid models wth coarse resolution are found to aritifically disperse NOx over rural areas, therefore overestimating rural levels of both NOx and O3. Regional net ozone production is too high in coarse grid models, because production of O3 is more efficient per molecule of NOx in the low-concentration regime of rural areas than in heavily polluted plumes from major emission sources. Ozone levels simulated by our model are shown to agree with observations in urban plumes and in rural regions. The model reproduces accurately average regional and peak ozone concentrations observed during a 4-day ozone episode. Computational costs for the model are reduced 25- to 100-fold as compared to fine-mesh models. ¿ American Geophysical Union 1990