Xray serves as the main diagnostic tool for variants of diseases but training experienced radiologist is time-consuming. So some machine learning approaches of classifying x-ray images have been proposed and tested. A recently released x-ray dataset – CheXpert – from Stanford Machine Learning lab which is much bigger than previously used dataset which can be served as better training and validating dataset. We use the CheXpert dataset and implemented the baseline model described in Jeremy Irvin’s paper. Then we will seek for any potential improvements mentioned in other papers and test on the dataset. Our main goal is to improve the baseline model or find another model which can exceed the baseline scores presented in Jeremy Irvin’s paper.