Input training patterns are presented to an autoassociative neural network, without desired responses. Learning is unsupervised. The training algorithm is the usual backpropagation which is a supervised learning algorithm. In an unusual manner, output desired responses are taken as the input patterns themselves. An autoassociative network is trained to produce output patterns which are identical to the input patterns that were in the original training patterns set. When presenting input patterns outside of the training set, the network outputs do not match the inputs. The difference between input and output is an error pattern. An input pattern in the training set will correspond with small errors while patterns not in the training set are associated with large errors. Once the network is trained, small errors indicate déjà vu, patterns that have been seen before.