Early brain tumor detection is crucial. Magnetic Resonance Imaging (MRI) provides clear images, and artificial intelligence (AI) with deep learning (DL) is revolutionizing automatic diagnosis, enhancing the speed and accuracy of medical image interpretation. In our study, we aimed to create a reliable method for classifying brain tumors in MRI scans using a technique called transfer learning. We used well-known deep learning models like XCEPTION, Res-Net50, InceptionResNetV2, and VGG16. These models were pretrained and then applied to pull out important features from brain MRI images. We conducted experiments on a dataset of 7300 images which contains three types of brain tumor (glioma, meningioma, pituitary). To ensure accurate and fast training, we first adjusted, processed, and enhanced the images. To evaluate how well our models performed, we used metrics like accuracy, sensitivity, precision, specificity, and F1-score. The results showed that our proposed CNN model, based on the ResNet50 architecture and using the ADAM optimizer, outperformed the other three models. The ResNet50 model demonstrated exceptional performance across accuracy, sensitivity, precision, specificity, and F1-score metrics in the dataset. In summary, our method stands out compared to existing research, suggesting its potential for quickly and accurately classifying brain tumors.