Computational fluid dynamics (CFD) is a mature method in wind engineering studies, but many air ventilation assessment (AVA) reports prepared by consulting firms in Hong Kong still suffer from the use of inappropriate techniques and computational settings in the CFD modeling stage. The main reason behind this is the lack of informative guidelines relating to CFD model settings in the AVA Technical Circular issued by local authorities. Other aspects of AVA that require improvement include terrain modeling techniques and understanding the sensitivity of modeled topographical size to the wind environment. This paper revisits and summarizes important aspects of current best-practice guidelines for robust CFD simulations. The study compares two conventional approaches to terrain treatment through a series of sensitivity tests. The first approach uses terrain features to cover the entire ground of the computational domain, whereas the second imitates wind tunnel experiments in which the domain includes an inclined buffer area. One drawback of the first approach was found to be the occurrence of numerical divergence when the terrain size is small; meanwhile, in the latter approach, the inclination angle of the buffer area must not exceed 30° to achieve robust results. The final stage of the study validates the results of CFD simulations based on the two approaches with experimental data from a dense city. The approach that mimics wind tunnel experiments with a buffer zone was found to achieve better correlations, and smaller normalized mean square errors with respect to the experimental data and is thus considered superior. •Best Practice Guidelines for CFD simulations are revisited.•Two techniques for terrain modeling via CFD are validated by wind tunnel experiment.•VR and TKE are insensitive to a buffer area with an inclination angle less than 30°.•Simulation results are influenced by terrain size and treatment techniques.