Shoe prints found on the crime scene are useful to understand the crime dynamics. Several semi-automatic and automatic systems have already been proposed to identify the make and model of the shoe that left the mark on the crime scene, however these systems have been tested on simulated shoe marks, i.e. not coming from real crime scene but artificially synthesized with different noise adding techniques. In this paper, we propose a novel descriptor, based on the Mahalanobis distance, to retrieve the footwear that produced the shoe mark. Differently from others, the performance test of the selected descriptor is done on shoe marks coming from real crime scenes, and the results are promising.