Vehicle classification is an important subject of study due to its significance in a number of areas including law enforcement, traffic surveillance, autonomous navigation, and transportation management. While numerous approaches have been proposed, few studies have been published with regard to the multi-view classification of vehicles captured in real surveillance. In this paper, we consider the multi-view classification of vehicles as an attribute prediction problem with views (rear, front, and side) as attributes. The corresponding multi-task learning is implemented in the Region-based Convolutional Neural Network (RCNN) framework, which classifies vehicle categories (car, truck, bus, and van) and predicts the attributes simultaneously. Experiments on a field-captured vehicle dataset provide satisfactory results, with approximate 83% accuracy for vehicle type classification and over 90% accuracy for attribute prediction.