A key factor in engineering integrated devices such as electro-optic switches or waveguides is the patterning of high quality crystalline thin films into specific geometries. In this contribution high vacuum chemical vapor deposition (HV-CVD) was employed to grow titanium dioxide (TiO2) patterns onto silicon. The directed nature of precursor transport - which originates from the high vacuum environment during the process - allows shading certain regions on the substrate by shadow masks and thus depositing patterned thin films. While the use of such masks is an emerging field in stencil or shadow mask lithography, their use for structuring thin films within HV-CVD has not been reported so far. The advantage of the employed technique is the precise control of lateral spacing and of the distance between shading mask and substrate surface which is achieved by manufacturing them directly on the substrate. As precursor transport takes place in the molecular flow regime, the precursor impinging rates (and therefore the film growth rates) on the surface can be simulated as function of the reactor and shading mask geometry using a comparatively simple mathematical model. In the current contribution such a mathematical model, which predicts impinging rates on plain or shadow mask structured substrates, is presented. Its validity is confirmed by TiO2-deposition on plain silicon substrates (450 degrees C) using titanium tetra isopropoxide as precursor. Limitations of the patterning process are investigated by the deposition of TiO2 on structured substrates and subsequent shadow mask lift-off. The geometry of the deposits is according to the mathematical model. Shading effects due to the growing film enables to fabricate deposits with predetermined variations in topography and non-flat top deposits which are complicated to obtain by classical clean room processes. As a result of the enhanced residual pressure of decomposition products and titanium precursors and the corresponding surface coverage of them, growth conditions differ in proximity of the shadowed areas compared to the case of plain substrates and the obtained film thickness differs significantly from predictions. (C) 2014 Elsevier B.V. All rights reserved.