Pneumonia is a severe lung infection that can be classified as bacterial and viral. Accurate and timely detection is essential for effective treatment planning and improved patient outcomes. Chest X-ray imaging is widely used for pneumonia diagnosis, offering detailed visual representations of lung structures. However, manual interpretation can be subjective and time-consuming, highlighting the need for automated diagnostic solutions. This study explores the application of deep learning models for pneumonia detection using a dataset of over 5,800 Chest X-ray images. We evaluate and compare three deep learning models: VGG-16, DenseNet-121, and EfficientNet-B3. Employing transfer learning with ImageNet pre-trained weights, we fine-tune each model for three-class classification and evaluate their performance. A comprehensive analysis is conducted to assess their strengths, limitations and performance for pneumonia detection. Experimental results show DenseNet-121 leading with 0.84 accuracy and balanced recalls (0.84 Normal, 0.97 Bacterial, 0.64 Viral), benefiting from its efficient 8 million parameters and dense connectivity. EfficientNet-B3 achieves 0.76 accuracy, excelling in Bacterial (0.97) and Viral (0.81) recalls, but lagging in Normal detection (0.52). VGG-16, with 0.71 accuracy, demonstrates high Bacterial (0.88) and Viral (0.80) recalls but poor Normal recall (0.47). The study highlights DenseNet-121’s superiority for clinical use, particularly for bacterial pneumonia, and suggests future enhancements through ensemble approaches or expanded datasets.