In the paper we address the problem of usefulness of colour image filtering before segmentation. Our interests are limited to non-linear colour filters working in the spatial domain. Most often comparing such filters is based on calculation of different quality factors (e.g. SNR, NCD etc.). The main idea of this paper is to use an evaluation function, coming from segmentation, to estimate the usefulness of preprocesing. The experiments were realized using both original and noisy images corrupted by Gaussian and impulsive noise.