Almost all prior work on modeling the dependence of acoustic emissions on tool wear have concentrated on the effect of wear-level on the sound. We give justification for including the wear-rate information contained in the sound to improve estimation of wear. A physically plausible model is proposed which results in a Hidden Markov Model (HMM) whose hidden states are the wear level and rate and observations are the feature vectors extracted from the sound. We also present an efficient method for picking feature vectors that are most useful for the classification problem.