A brain tumor is an abnormal growth of cells within the brain or spinal canal, which can be either benign or malignant. These tumors can disrupt brain function, leading to symptoms such as headaches, seizures, and cognitive issues. Therefore, early detection of tumors from Magnetic Resonance Imaging(MRI) images is crucial. Despite advancements in deep learning,most existing studies face overfitting issues in deep learning models, limiting their performance. Additionally, the lack of validation on multiple datasets raises concerns about the models’ generalizability. To address this gap, we propose two hybrid models, VGG16-LSTM and VGG19-LSTM, which have proven efficient in classifying brain tumors from MRI images using two publicly available datasets. We employed several preprocessing techniques, including image resizing and data augmentation, and meticulously discussed the hybrid model architectures. In addition, we used five pretrained Convolution Neural Network(CNN) models—VGG16, VGG19, InceptionV3, ResNet50, and Xception—for comparison by leveraging the transfer learning techniques. VGG16-LSTM achieved the highest accuracy of 91.00% on the Fagshire dataset, on the other hand VGG16 showed the highest accuracy of 98.17% on the BR35H dataset, slightly outperforming VGG16-LSTM and VGG19-LSTM, which achieved 97.5% and 97.67%, respectively. To gain further insights, we calculated precision, recall, F1-score, and specificity. The hybrid models generally outperformed the pretrained CNN models in most performance metrics and demonstrated less overfitting on these datasets. We believe this research paves the way for further advancements in the field.