Analysing brain magnetic resonance angiography (MRA) images is important for detecting arteriovenous malformations and aneurysms. To detect these diseases, extracting the vessel structure in the image can be seen as a first step. In this paper, it was aimed to classify the cubic image parts obtained from brain MRA images according to whether they belong to vein structure or not. For this purpose, a 9 layers deep convolutional neural network (CNN) architecture is designed. With the model trained using this architecture, 85% accuracy was obtained in the classification performed on the test data.