Technologies of computer-aided diagnosis（CAD）have been applied to the detection or differential diagnosis of lesions so far. However, there are few cases where CAD technology is applied to therapy. The purpose of this study is to propose a method for the prognostic prediction of glioblastomas by using genes and image features. Our database consists of messenger RNAs（mRNAs）, MR images and survival times obtained from 62 patients（41 males, 21 females）.First,we selected 20 genes from 12042 mRNA, which were having small Deviance residuals in the univariate Cox regression. After the tumor regions in MR images were manually segmented, we calculated 94 image features such as size, shape, pixel value,and texture in the segmented tumor regions. Twenty image features were selected by the same manner used in the selection of genes. We estimated the survivor functions for 62 patients by using Cox regression with selected genes and image features. The outputs of Cox regression were evaluated by using the time-dependent receiver operating characteristic （ROC）analysis. The experimental result showed that the time-specific area under the curve AUC（t）exceeded 0.82 when the survival time less than 296 days. Because our method can provide objective information about the patient's prognosis, it would be useful for assisting in the determination of suitable treatment of brain tumor patients.