A recognition system consisting of attributed grammars and a two-layer perceptron for classifying brain-stem auditory evoked potentials is described. An evoked potential waveform is filtered and converted to a string of terminal symbols. This string is then processed by a regular attributed grammar. Its semantic functions return a list of numeric features that are further processed by a two-layer perceptron. The paper discusses the results of a large number of experiments that have been conducted to obtain the performance characteristics of a multilayer perceptron on a specified pattern recognition problem.