3D feature descriptors extracted from point clouds are becoming a promising information source for many applications. These include object/shape recognition, building information and civil structures modeling, autonomous navigation systems, etc. Considering these trends, this paper presents a classification system for vehicles based on the bag of visual words framework. The former extracts feature descriptors from range images being captured from a SICK LMS200 sensor. Our approach uses also visual information to estimate the vehicle velocity using a Kalman filter. The velocity estimation is used to properly register laser scans and build the scene point cloud. In this work, a dataset was set up by including the vehicle point cloud, related visual information, vehicle velocity estimation as well as captured label classes. Using this dataset, various 3D descriptors were tested and for the classification process a bag of visual words was employed while KD-trees were used to speed up the process. As a result, our approach can classify up to nine different classes of vehicles. In this work, the classifier performance was measured using Precision-Recall curves