SUMARY: Nowadays, users choose footwear based only in length and width measurements, but it is widely accepted that three-dimensional shape of the foot can help in good shoe fitting. However, both acquisition and processing of 3D data involves some technical and methodological difficulties. The combination of 3D scanning systems with mathematical classification techniques makes possible the development of best-fitting systems, which can help in the selection of shoes for a given customer. In this paper, a new approach for customized classification (assignment) of comfortable footwear is proposed. It uses 3D user's foot data (around 20.000 points), acquired with the hand held laser scanner FastScan®. The scanning procedure was improved by using automatic techniques to avoid its dependence on manually detected landmarks. Footwear fitting is predicted from scanned foot data by application of a statistical prediction model. The proposed approach was tested on a database of women footwear with good results. This approach closely matches the subjective assessment of footwear fit given by users and experts. sur une chaussure avec l'approche proposée est très proche de la valorisation subjective donnée par les sujets d'essais.