In this paper, the use of low-power wireless sensor networks for the classification of signals is described. The objective is to accurately classify the target signals using a novel decentralized approach that aims to reduce the energy consumption of the network nodes as much as possible. Since, often, the most consuming part of such nodes is the radio module, the proposed solution performs the classification of the selected features of a detected signal directly on the board, thus transmitting only the classification results. As a case study, the identification of vehicles that emit acoustic signals is considered. For this case, the process of feature extraction and selection is based on a spectral analysis of the signals, whereas the classification is carried out by support vector machines, which were chosen for their flexibility in classifying patterns. In particular, the use of the -SVM classification algorithm is proposed because it is expected to provide a good performance in terms of both implementation cost and classification accuracy. Finally, the advantages of the proposed solution in terms of energy consumption are illustrated with reference to an implementation based on the Mica2 sensor node by Crossbow Technology Inc.