Surface light scattering enables contactless and fast measurements of surface roughness. A surface inclination alters the direction of the scattering beam and thus the measured surface roughness is calculated from the detected intensity distribution. Hence, an accurate sensor–surface alignment is necessary. In order to achieve tilt-independent roughness measurements, a model-based evaluation approach for polychromatic speckle patterns is presented. By evaluating the shape of the superposed speckles, which occur for polychromatic illumination, with regard to the distance to the scattering centre, surfaces with an Sa roughness value in the range of 0.8–3.2 μm are measurable. Experimental investigations demonstrate that the influence of a surface tilt up to ± 1.25° on the roughness measurement is reduced by 90%. As a result, the robustness of the polychromatic speckle roughness measurement is improved, which allows to speed up the adjustment of the measurement system or the surface sample, respectively.