The generation of forces in the contact area between wheel and rail due to roughness is an inherently non-linear process. In order to provide an excitation input to existing linear analytical models of wheel/rail noise, we have examined two alternate calculation procedures for estimating these interaction forces: the distributed point reacting spring (DPRS) procedure and the Boussinesq procedure. Both procedures are based on estimating the blocked force produced by the roughness. The blocked force can then be used in the linear wheel/rail analytical models to calculate wheel and rail response and sound radiation. The DPRS procedure is substantially more computationally efficient than the Boussinesq and is found to provide similar estimates of the normal stress distribution in the contact zone for roughness wavelengths that are long compared to the contact area dimensions. On the other hand, the two procedures provide quite different estimates of the normal stress distribution in the contact zone when the roughness wavelengths are short compared to the dimensions of the contact area. When the two models are used to estimate the blocked interaction force between the wheel and rail for measured wheel and rail two-dimensional roughness distributions, however, the blocked force predictions are very similar even for wavelengths that are the order of or slightly shorter than the dimensions of the contact area.