COVID-19 quickly became a global pandemic after only four months of its first\ndetection. It is crucial to detect this disease as soon as possible to decrease\nits spread. The use of chest X-ray (CXR) images became an effective screening\nstrategy, complementary to the reverse transcription-polymerase chain reaction\n(RT-PCR). Convolutional neural networks (CNNs) are often used for automatic\nimage classification and they can be very useful in CXR diagnostics. In this\npaper, 21 different CNN architectures are tested and compared in the task of\nidentifying COVID-19 in CXR images. They were applied to the COVIDx8B dataset,\na large COVID-19 dataset with 16,352 CXR images coming from patients of at\nleast 51 countries. Ensembles of CNNs were also employed and they showed better\nefficacy than individual instances. The best individual CNN instance results\nwere achieved by DenseNet169, with an accuracy of 98.15% and an F1 score of\n98.12%. These were further increased to 99.25% and 99.24%, respectively,\nthrough an ensemble with five instances of DenseNet169. These results are\nhigher than those obtained in recent works using the same dataset.\n