Computational fluid dynamics (CFD) is now widely used as an essential tool in the development of automotive aerodynamics. However, the time required for repairing non-watertight geometries has recently become a serious problem in current CFD processes. Therefore, we developed an efficient simulation method that allows the flow around a non-watertight geometry to be computed on a Cartesian grid. This method can substantially reduce the turnaround time and effort required for CFD processes, because the repair work can be eliminated. The proposed method adopts an embedded boundary condition technique to capture arbitrary shapes more accurately on the background Cartesian grid. In addition, a local mesh refinement technique enables higher computational efficiency to be realized, and large-eddy simulation (LES) is used to reproduce high-Reynolds-number turbulent flow. Preliminary tests were performed on an engine bay configuration that had non-watertight geometries and a 1/5-scale model of an automobile configuration. As a result, the proposed method was confirmed to enable rapid grid generation and flow simulation around non-watertight geometries. Moreover, the computed results showed good agreement with experimental data.