COVID-19 has severely impacted healthcare systems and economies worldwide since its onset in late 2019. Rapid and accurate diagnosis is vital to control the spread. The golden standard for testing is reverse transcription polymerase chain reaction (RT-PCR), yet it has drawbacks. As an alternative, chest radiography-based diagnosis presented results near to the RT-PCR. The study proposes a Transfer Learning(TL)-based approach for classifying images of chest X-ray into normal, COVID-19, and pneumonia categories, using data from two publicly available Kaggle datasets. After the preprocessing, seven pretrained Convolutional Neural Networks (CNNs) including ResNet50, ResNet101, VGG16, VGG19, InceptionV3, MobileNet and Xception are fine-tuned by adding new fully connected layers. MobileNet achieved best accuracy of 96.21% on one dataset while ResNet50 attained 94.86% on the second dataset. High precision, recall and F1-scores are also obtained. The consistent performance across CNN architectures demonstrates the effectiveness of TL in COVID-19 detection from chest radiographs, presenting a rapid and reliable solution for diagnosis.