With the outbreak of new coronary pneumonia, a variety of pneumonia diseases are emerging, doctors often have misdiagnoses and omissions. To assist doctors in better clinical diagnosis, a variety of high-performance neural network structures have been applied to the X-ray recognition of pneumonia, and the X-ray image recognition accuracy will be improved, will effectively improve the efficiency of the hospital detection, at the same time, can also avoid the patient to miss the best rescue time. To this end, this paper compares and analyses the recognition performance of the three currently used neural network structures in the recognition of pneumonia X-ray images, uses some publicly available datasets and a mixed set composed of different datasets to conduct experiments, and summarises the advantages of the models and the direction of possible improvement. The three neural network models are presented, compared, and analysed to suggest useful references for the recognition of pneumonia X-ray images. Finally, an analysis and outlook for future neural network models for pneumonia x-ray image recognition is presented.