Patient MGMT (O 6 methylguanine DNA methyltransferase) status has been identified essential for the responsiveness to chemotherapy in glioblastoma patients and therefore depicts an important clinical factor. Testing for MGMT methylation is invasive, time consuming and costly and lacks a uniform gold standard. We studied MGMT status assessment by multi-parametric magnetic resonance imaging (mpMRI) scans and tested the ability of deep learning for classification of this task. To overcome the limited number of training examples we used a transfer learning approach based on the video clip classification network C3D [30], allowing for full exploitation of three dimensional information in the MR images. MRI sequences were fused using a locally connected layer. Our approach was able to differentiate MGMT methylated from unmethylated patients with an area under the receiver operating characteristics curve (AUC) of 0.689 for the public validation set. On the private test set AUC was given by 0.577. Further studies for assessment of clinical importance and predictive power in terms of survival are needed.