The outbreak of the novel coronavirus brought the world to a halt, affecting the health care system worldwide. Thus, the need for a rapid and accurate detection method emerged to help stop the high death rates. The use of convolutional neural networks with chest X-ray screening has been demonstrated to be effective in early COVID-19 diagnosis. Therefore, in this study, four distinct transfer learning models were investigated to detect COVID-19 in two-class (case 1) and three-class (case 2) classifications. In both cases, the classifications were carried out on a balanced dataset. In case 1, a binary classification was performed between COVID-19 patients and non-infected X-rays, while in case 2, a multi-classification between COVID-19, viral pneumonia patients, and non-infected X-rays was presented. The confusion matrices obtained from each model evaluated the models' performance. The test results are presented in terms of accuracy, precision, recall, and F1-Score. The results demonstrated that the visual geometry group–16 model had the highest accuracy in both scenarios compared to other models, with a binary classification accuracy of up to 99% and a multi-classification accuracy of up to 94%.