Structural responses obtained with finite element simulations normally differ from those measured on physical prototypes. In the case of monolithic structures, the differences between the simulated and measured responses are mainly caused by inaccuracies in the geometry and material behavior. The present work focuses on evaluating the impact of using a high-fidelity representation of the actual geometry on the differences between measured and computed resonant frequencies and mode shapes. This paper presents a study that was performed on a cast-iron lantern housing of a gear box. In a first step, the resonant frequencies and modes shapes of the test structure were measured using impact testing. Next, optical scanning and photogrammetric techniques were used to obtain a 3D virtual point cloud model which accurately describes the surface of the lantern housing. This point cloud was then used to generate a 3D solid finite element model representing the as-built geometry of the housing. To evaluate the impact of using the actual geometry on the correlation and model updating results, two FE-models were used: an FE-model derived from the measured geometry and an FE-model derived from the CAD model of the lantern housing. Both models have a similar mesh density and mesh quality. These two models were first correlated with the measured modal data and then updated. The geometry appeared to have a significant impact on both the correlation and updating results.