Abstract Flow and pollutant dispersion in a densely built-up area of Seoul, Korea, are numerically examined using a computational fluid dynamics (CFD) model coupled to a mesoscale model [fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5)]. The CFD model used is a Reynolds-averaged Navier–Stokes equations model with the renormalization group k − ɛ turbulence model. A one-way nesting method is employed in this study. MM5-simulated data are linearly interpolated in time and space to provide time-dependent boundary conditions for the CFD model integration. In the MM5 simulation, four one-way nested computational domains are considered, and the innermost domain with a horizontal grid size of 1 km covers the Seoul metropolitan area and its adjacent areas, including a part of the Yellow Sea. The NCEP final analysis data are used as initial and boundary conditions for MM5. MM5 is integrated for 48 h starting from 0300 LST 1 June 2004 and the coupled CFD–MM5 model is integrated for 24 h starting from 0300 LST 2 June 2004. During the two-day period, a high-pressure system was dominant over the Korean peninsula, with clear conditions and weak synoptic winds. MM5 simulates local circulations characterized by sea breezes and mountain/valley winds. MM5-simulated synoptic weather and near-surface temperatures and winds are well matched with the observed ones. Results from the coupled CFD–MM5 model simulation show that the flow in the presence of real building clusters can change significantly as the ambient wind speed and direction change. Diurnally varying local circulations mainly cause changes in ambient wind speed and direction in the present simulation. Some characteristic flows—such as the double-eddy circulation, channeling flow, and vertical recirculation vortex—are simulated. Pollutant dispersion pattern and the degree of lateral pollutant dispersion are shown to be complicated in the presence of real building clusters and under varying ambient wind speed and direction. This study suggests that because of the sensitive dependency of urban flow and pollutant dispersion on variations in ambient wind, time-dependent boundary conditions should be used to better simulate or predict them when the ambient wind varies over the period of CFD model simulation.