In Glioblastoma Multiforme (GBM) image-derived features (“radiomics”) could help in individualizing patient management. Simple geometric features of tumors (necrosis, edema, active tumor) and first-order statistics in Magnetic Resonance Imaging (MRI) are used in clinical practice. However, these features provide limited characterization power because they do not incorporate spatial information and thus cannot differentiate patterns. The aim of this work is to develop and evaluate a methodological framework dedicated to building a prognostic model based on heterogeneity textural features of multimodal MRI sequences (T1, T1-contrast, T2 and FLAIR) in GBM. The proposed workflow consists in i) registering the available 3D multimodal MR images and segmenting the tumor volume, ii) extracting image features such as heterogeneity metrics and iii) building a prognostic model by selecting, ranking and combining optimal features through machine learning (Support Vector Machine). This framework was applied to 40 histologically proven GBM patients with the endpoint being overall survival (OS) classified as above or below the median survival (15 months). The models combining features from a maximum of two modalities were evaluated using leave-one-out cross-validation (LOOCV). A classification accuracy of 90% (sensitivity 85%, specificity 95%) was obtained by combining features from T1 pre-contrast and T1 post-contrast sequences. Our results suggest that several textural features in each MR sequence have prognostic value in GBM.