This paper introduces a new approach for the generation of as-scanned point clouds of CAD assembly models. The resulting point clouds incorporate various realistic artifacts that would appear if the corresponding real objects were digitalized with a laser scanner. Such a virtual Reverse Engineering technique can produce a huge amount of realistic point clouds much faster than using classical time-consuming Reverse Engineering techniques on real physical objects. Here, there is no need to use a laser scanner and the post-processing steps are automatic. Using this technique, it is easy to create large databases of point clouds automatically segmented and labeled from the CAD models and which can be used for supervised machine learning. The proposed approach starts by generating a triangle mesh wrapping the CAD assembly model to be reverse engineered. The resulting watertight mesh is then sampled to obtain a more realistic distribution of points. The occlusion phenomenon is then simulated using a hidden point removal algorithm launched from several viewpoints. A misalignment procedure can optionally be used to simulate the fact that in real-life Reverse Engineering the position and orientation of the laser scanner and/or real object would have been changed to get a different scanning viewpoint. The virtual Reverse Engineering process ends by generating noise and by inserting outliers. The approach is illustrated and validated on several industrial examples.