In this paper, we propose a neuro-classification method of the new and used bills using time-series acoustic data. The technique used here is based on an extension of an adaptive digital filter (ADF) by Widrow and the error back-propagation method. Two-stage ADFs are used to detect the desired acoustic data of bill from noisy input data. In the first stage, superfluous signals are eliminated from input signals and in the next stage, only the desired acoustic data is detected from output signal of the two-stage ADFs. The output signal of two-stage ADFs is transformed into spectral data to produce an input pattern to a neural network (NN). The NN is used to discriminate the new and used bills. It is shown that the experimental result using two-stage ADFs is better than that obtained by using original observation data.