Detecting brain tumors from MRI images is a critical task in medical diagnosis, but it often requires a large amount of labeled data, which is difficult to obtain, especially for rare conditions. To address this challenge, this study investigates Self-Supervised Learning (SSL) methods that can learn from unlabeled data. Three SSL models are applied and compared: SimCLR, Bootstrap Your Own Latent (BYOL), and Masked Autoencoders (MAE), for classifying brain MRI images into four categories: glioma, meningioma, pituitary tumor, and no-tumor. The models are first pre-trained without labels and then fine-tuned for the classification task. The results indicate that MAE performs the best, achieving high F1-scores for all tumor types, including 0.96 for glioma and 1.00 for no-tumor cases. The findings demonstrate that SSL is a promising approach for medical image classification in situations where labeled data is limited.