Single-sided and cross- ventilation of a cube-shaped building at various incident wind directions were stimulated using computational fluid dynamics (CFD). The simulation used either the Reynolds-averaged Navier-Stokes (RANS) equations in conjunction with a 2-equation turbulence model (Standard k-ε model (SKT), Realizable k-ε model (RLZ), or Renormalization group k-ε model (RNG)), or Large Eddy Simulation (LES) with the Wall-Adapting Local Eddy-viscosity model (WALE). LES showed good agreement with wind tunnel data when modeling indoor and outdoor wind fields; RLZ had slightly better results than SKT and RNG. Two evaluation methods were examined: the integration of opening velocities method under-estimated single-sided ventilation rates but over-estimated cross-ventilation rates; the tracer-gas decay method was more computationally demanding of the two. When the tracer-gas method and LES were used in combination, their estimations were most accurate especially if the ventilation was driven by wind flow fluctuations. RANS and LES predicted steady decreases of 92.5% and 81.8% (single-sided ventilation), and 52.6% and 37.2% (cross-ventilation) in ventilation rate as incident wind direction varies from 0° to 90°. Discrepancies between the RANS and LES predictions of ventilation rates were mainly attributable to their respective turbulence-modeling methods.