This paper proposed a support vector machine (SVM) based classification method to identify diversified pathological voices. Sound signals were sampled from the pronunciation of a vowel "a" vocalized by 214 subjects, including 181 patients suffered from various dysphonias (such as polypoid degeneration, adductor spasmodic dysphonia, vocal fatigue, vocal tremor, vocal fold edema, hyperfunction, and erythema), and 33 healthy subjects. 25 acoustic parameters were calculated from the sampled data for each subject. The original acoustic dataset was first transformed via principal component analysis (PCA) method into a new feature space. To learn the identification boundary for healthy and pathological voices, a soft-margin SVM and three kinds of kernels were examined. The results under different combination of parameters and kernels were investigated. The effectiveness of SVM-based approach seems to be promising in the application of pathological voice identification.