In this paper, the acoustic parameters of voice are studied, and the differences between normal voice and pathological voice are analyzed. 53 samples of normal voice and 173 samples of pathological voice are selected from the MEEL database of Kay Company for experiments. Bayesian network, support vector machine, K-nearest neighbor, C4.5 decision tree and random forest are used to recognize the pathological voices. The experimental results show that the acoustic characteristic parameters have distinguish ability, and the average recognition rate is over 86%. The pathological voice recognition rate based on random forest classifier is the best, reaching the rate of 90.27%.