Abstract: In order to do some researches on the pathological voice classification in the aspect of acoustic mechanism, a dynamics model parameter inversion method is proposed. Based on vocal fold structure and Bernoulli's law, a mechanical model of vocal fold called two-mass model is established, and then coupled to the glottal airflow to produce glottal excitation. Simultaneously, a set of model parameters needed to be optimized is selected. Finally, a model parameter inversion procedure is to reproduce the glottal excitation that will match with the objective obtained from the vocal voice using iterative adaptive inverse filtering(IAIF). Within the inversion procedure, a parameter optimization is performed using genetic algorithm(GA). That the glottal excitation time-frequency domain features' matching relative error is less than 2% shows the good performance of the inversion procedure. The average normalized scaling on the basis of the optimized parameters sets is proposed to overcome vocal fold asymmetry's defects in pathological voice classification,realizing the valid and entire distinguishment.