This paper provides a comparison of the performance of five different neural network architectures in diagnosing machinery faults. The network architectures include perceptrons, linear filters, feed-forward, self-organizing, and LVQ. The study provides a critical analysis of the performance of each network on a test rig with different faults. The comparison discusses the success rate in network training and identification of faults including: unbalance and looseness. It is shown that the perceptron and LVQ architectures were superior and achieved 100% diagnosis on the cases presented.