The feasibility of using an artificial network for identifying faults in induction motors has been demonstrated previously by the authors. In this work, the network was used as a learning and pattern recognition device, and was able to successfully associate input signal patterns with appropriate machine states. The neural network used was the multilayered perceptron (MLP), trained by a backpropagation algorithm. However, MLP lacks flexibility since it requires fully labelled input-output pairs (i.e. training of the network is supervised). This limitation can be removed by the use of an alternative approach, using unsupervised methods, such as the Kohonen feature maps (KFM) technique. The results of applying KFM to condition monitoring of electrical drives are reported in this paper, and they reveal the practical advantages of unsupervised systems, which include the ability to learn and produce classifications without supervision. Because of the natural parallel architecture of neural networks, they are also ideally suited to the use of multiple transducer inputs, which can greatly enhance the reliability of decisions made regarding the state of machine performance or condition.