We report on a novel method of acoustic wear analysis using spectral classification based on a model of mammalian audition. This approach uses biologically-inspired pre-processing filters that give a multi- resolution representation of the sound timbre. A Tree Structured Vector Quantizer (TSVQ) is used to create a classification tree based on the wear labels. We have obtained encouraging results for tools of different diameters, cutting different materials. Trees trained on one kind of data seem to generalize well to new data sets.