Abstract: A novel nonlinear noise canceller based on additive-multiplicative fuzzy neural network (AMFNN) was proposed, and its model, learning algorithm and universal approximationwere discussed. In the cancellation process, AMFNN is used to approximate the noises, and the noises are cancelled to recover the useful signals. The new canceller has the advantages of neural networks, such as the ability of parallel distributed information processing, fault-tolerance and robustness, and is easy to use as a noise canceller.