We present an image classifier based on the CheXNet and a transfer learning stage to classify chest X-Ray images according to three labels: COVID-19, viral pneumonia and normal. CheXNet is a DenseNet121 that has been trained twice, firstly on ImageNet and then, for classification of pneumonia and other 13 chest diseases, over a large chest X-Ray database (ChestX- ray14). The proposed network reached a test accuracy of 97.8% and, for the COVID-19 class, of 98.3%. In order to clarify the modus operandi of the network, we used Layer Wise Relevance Propagation (LRP) to generate heat maps, indicating an analytical path for future research on diagnosis.