Improving the accuracy and reproducibility during patient positioning is of paramount importance. Hence, the goal of this work is to characterize the aspects of image blurring occurring during carbon target bremsstrahlung portal imaging and to assess the applicability of a deconvolution algorithm. Blurring effects involved in this method of portal imaging are electron scattering inside the EPID, geometric blurring due to the photon source size and photon scattering inside the patient. These effects can all be described by convolutions using as the convolutional kernel a Lorentz function, whose FWHM is 2λ. The λ values measured for these effects range from 0.2 mm to 0.45 mm, and an iterative 2D-deconvolution of carbon target portal images was performed accordingly. A significant decrease in the image blurring of test objects has been achieved and confirmed by analyzing the RMTF. However for clinical images, the deconvolution method is presently faced with the problem of the associated increase of image noise.