Conventional numerical simulations are hard-pressed to give the details of air flow in the whole urban area because of the limitation of computing resources, and this challenge can be effectively resolved when urban building complexes are modelled by the porous media model (PMM). However, the resistance coefficient (Cf) needed to run PMM has not been determined so far. Therefore, an anisotropic prediction model of Cf is developed for simulating wind flow through building complexes, and a wind direction index (aθ) is defined to characterize the anisotropy of Cf. Significance tests are conducted to analyze the influence of building configuration, approaching wind velocity and aθ on Cf. It is found that Cf depends mainly on building area density (λp) and aθ, so a model related to λp and aθ for predicting Cf is obtained by regression analysis. The feasibility of Cf within PMM is validated by results from the fully resolved building model (FRBM). The results show that PMM predicts the macroscopic airflow through the building array generally well except for streamwise velocity near the ground and at mid-height of the array, especially the macroscopic variations of vertical velocity and pressure simulated by PMM agree well with FRBM. PMM can also predict the macroscopic turbulence energy well for most of the regions inside the building array. Moreover, the relative differences between PMM and FRBM are about 2.8% for blocking effect on airflow and 1.5% for the ventilation capacity within the array, which are measured by pressure loss and in-canopy velocity, respectively.