On the basis of prostate tumor diagnosis by nuclear Magnetic Resonance Imaging(MRI), a two-stage ensemble Support Vector Machine(SVM) method were proposed to realize the prostate tumor aided diagnosis. Firstly, the statistical features, invariant moment features and the texture feature of the Area of Interest( ROI )for the prostate in a MRI image were extracted. Then, SVM parameters were disturbed by using different kernel functions in different feature spaces, and the first ensemble was carried out by relative majority voting. Furthermore, the results of first ensemble were integrated again by the relative majority voting. Finally, MRI images of prostate patients were regarded as original data, and two-stage ensemble SVM were utilized to aid tumor diagnosis. Experiment results show that the classification accuracy from the first ensemble has improved by 26.67% as compared with that of single-stage SVM and that from the second ensemble has improved 3.33% than that of the first ensemble. These results illustrate that the proposed algorithm can improve the recognition accuracy of prostate tumor effectively.