In this paper, we present a non-supervised method- ology for edge detection in color images based on belief functions and their combination. Our algorithm is based on the fusion of local edge detectors results expressed into basic belief assignments thanks to a flexible modeling, and the proportional conflict redis- tribution rule developed in DSmT framework. The application of this new belief-based edge detector is tested both on original (noise-free) Lena's picture and on a modified image including artificial pixel noises to show the ability of our algorithm to work on noisy images too.