A method combined probabilistic principal components analysis(PPCA) with support vector machine(SVM) was presented for analyzing SELDI-TOF-MS data generated from clinical proteomic study.Using PPCA for feature extraction on 322 MS data set,225 MS data set were randomly selected as learning set for establish SVM model，and the remaining 97 data set were selected as a testing set for prediction and verification.Root mean square error,recognition rate and predictive rate were used to evaluate the model′s classification performance,respectively.To verify the PPCA-SVM model′s classification performance further,the proposed model with partial least squares (PLS) model and PCA-SVM model were compared.The results showed that the recognition rates for PLS,PCA-SVM and PPCA-SVM were 90.92%, 99.23%and 99.01%,respectively,the predictive rates for PLS,PCA-SVM and PPCA-SVM were 76.38%,84.63% and 90.41%,respetively.Experimental results showed that proposed PPCA-SVM model was an accurate and repeatable method for automatically detecting prostate cancer.The method provides a new approach for early diagnosis of prostate cancer in clinic.