The growing computational capacity available at cheap prices is making FE numerical simulation of metal forming processes a key factor in process design and control, allowing reducing prototyping and downtime of costs. In order to get accurate simulation results, the critical aspect in commercially available FE codes is the proper model calibration. This aspect is of even greater importance in simulating special processes, which need non-standard settings in order to give effective process representation. An example of this is the flow-forming process: the complex tools movement, the localized strains and the high speed rigid rotation of the part require the use of ALE (Arbitrary Langrangian-Eularian) mesh in order to reduce computation times. This work concerns the calibration of a light alloy car rim flow-forming process simulation performed with Transvalor Forge®. The detachment at the mandrel-part interface due to material backflow was highlighted by preliminary simulations as one of the main issues deriving by inaccurate model calibration. A sensitivity analysis on the main simulation settings variables showed that part-tools contact conditions and material rhelogical characteristics have the heavier influence on simulation results, while frictional and heat transfer conditions effect is less important. The contact conditions, on the one hand, are bound with the part tangential discretization and the integration time step definition which are in turn linked with the ALE mesh parameters; optimal mesh settings were found as a trade-off solutions between simulation accuracy and computation time required. The rheological characteristics, on the other hand, have a considerable effect on the material backflow behavior; an iterative optimization was launched in order to find the material Spittel’s model constants which reduce the backflow. The results were critically assessed by comparing the resulting flow curves with those experimentally determined by means of compression tests. The use of a more complex rheological model was finally evaluated.