A brain tumor is an accumulation of malignant cells that results from unrestrained cell division. Tumors can result in crucial effects if they are not promptly and accurately recognized. Misdiagnosis can result in ineffective therapy, which decreases the patient's survival rate. The standard procedure for determining the presence of brain tumors and the type of tumors is magnetic resonance imaging (MRI). But as technology advances, it gets harder to comprehend huge amounts of data generated in an acceptable time. However, building a deep learning model from the start requires collecting enormous amounts of labeled data, which is a costly, time-consuming operation. A method to solve these issues is transfer learning of a deep learning model that has already been trained on the ImageNet dataset. In this research, the classification of brain tumors using several pre-trained deep learning models, i.e., different variations of ResNet, VGG, and DenseNet models, are being trained on a brain tumor dataset and compared. According to experiments, the ResNet50 model with a fine-tuned and transfer learning approach has achieved the highest training accuracy of 99%, validation accuracy of 96%, and test accuracy of 80%.