Micro-SWIFT SPRAY (MSS) is a 3D Lagrangian particle dispersion model that maintains a good balance between accuracy and computational cost. However, its capabilities for air dispersion modeling in the presence of both complex topography and high building densities have not been investigated for nuclear emergency response. In this study, MSS is systematically evaluated against two wind tunnel experiments that simulate a typical Chinese nuclear power plant with the above two features. The MSS predictions are compared with both 2D horizontal and vertical measurements. Sensitivity studies are performed with respect to the particle number, the lower bound of the turbulence intensity, and the horizontal and vertical grid size. The results demonstrate that ground-level predictions of both wind and radionuclide concentrations are in satisfactory agreement with the measurements under optimized parameter values. The vertical predictions exhibit site-dependent accuracy, but generally consistent tendencies. The default lower bound of the turbulence intensity in MSS may be insufficient for reproducing the correct plume width observed in the wind tunnel experiments. An increased lower bound is suggested to solve this problem. In addition, artificially high concentrations may arise near steep slopes if large horizontal/vertical grid sizes are used. Suitable parameters for preventing this problem are also provided.