Brain cancer is a complex disease, and it is increasing rapidly. Although the incidence rate of brain tumours is lower than other cancers, it is still the most serious disease threatening human lives. For effective treatment, detecting and diagnosing brain tumours are essential by an accurate and quick method. Though there has been an interest in using pattern recognition techniques to classify and grade tumours from Magnetic Resonance Imaging (MRI) images, effective and accurate grading remains difficult and subjective. This paper employs deep learning with the inception concept to classify the MRI brain images. A simple architecture is designed with convolution layers, max-pooling layers for feature extraction, and a fully connected layer for classification. 200 MRI images from the Repository of Molecular Brain Neoplasia Data (REMBRANDT) are used. Experimental results show that the proposed architecture obtains the best accuracy of 98% with a 0.01 learning rate and 20 epochs.