Railway noise contributes to environmental noise and may be a nuisance for people living near urban or interurban railway lines. Rolling noise is the main noise source over a wide speed range, resulting from the roughness of the railhead and wheel treads. The combined roughness at the wheel/rail contact zone produces vibrations that excite the track and the wheels, which then radiate noise. Maintaining a correct surface quality is a concern of rail network managers, requiring a knowledge of their network condition on large scales to optimise costly corrective actions. Indirect measurement methods, based on on-board vibration or acoustic sensors on a train set and combined with an inverse approach, allow the combined roughness to be estimated from the vibro-acoustic quantities. Then, removing the wheel roughness contribution is essential for properly assessing the rail condition. This can be achieved either by using smooth wheels, which is a stringent solution, or by implementing appropriate signal processing approaches taking advantage of the respective periodic or random properties of wheel and rail roughness. This paper reviews separation approaches available in the scientific literature. It also emphasizes the effect of the lateral shift of the contact point on the wheel tread, interfering with the deterministic nature of the wheel roughness contribution. This work is part of the French project MEEQUAI developing an on-board measurement system for the assessment of rail roughness.