Pneumonia is a severe illness, particularly affecting infants, young children, individuals above 65 years old, and those with compromised health or weakened immune systems. Pneumonia is a dangerous disease and often causes death if not being detected and treated instantly. Extensive research has revealed various pathogens, including bacteria, viruses, and fungi, as the potential causes of pneumonia. Furthermore, the global spread of COVID-19 has had devastating impacts on the global economy and public health. Therefore, an accurate machine learning-based application for pneumonia diagnosis would significantly save time, resources, and enable timely treatment, reducing the risk of complications. This study proposes a transfer learning approach for pneumonia classification. Specifically, this work has utilized the pre-trained model (e.g., the VGG16 model) which already had very good parameters on large dataset. Based on the pre-trained model, we removed the last layer and replaced it with new fully-connected layers and an output layer to fit with problem of pneumonia classification, re-trained and fine-turned the model to classify pneumonia diseases. We collect X-ray images from a variety of data sources to build a dataset for three classes such as Normal, COVID-19, and Viral diseases. Experimental results on a dataset of 2500 X-ray images show that using transfer learning approach can improve the accuracy of the prediction model.