In GNSS tomography, precise information about the tropospheric water vapor distribution is derived from integral measurements like GNSS slant wet delays (SWDs). Therefore, the functional relation between observations and unknowns, i.e. the signal paths have to be accurately known for each station-satellite pair involved. Since GNSS signals are prone to atmospheric bending effects, a straight line assumption is not sufficient for elevation angles < 15°. Thus, in the following, a mixed 2D piecewise linear ray-tracing approach is introduced and possible error sources in reconstruction of the bended signal paths are analyzed in more detail. Especially, if low elevation observations (ε < 10°) are considered, unmodeled bending effects can introduce a systematic error of up to 10–20 ppm, on average of 1–2 ppm into the tomography solution. Thereby, the ray-tracing approach itself but primarily the quality of the a priori field has a significant impact on the reconstruction quality. In order to overcome possible limitations in the a priori field, an iterative tomography processing strategy is applied. Further, a bending model helps to reduce the number of processing steps by up to 85 %. In consequence, the developed mixed ray-tracing approach allows not only for a correct treatment of low elevation observations but is also fast and applicable for near real-time applications.