The thyroid gland’s significance to the human body makes thyroid cancer prediction vital. By a proportion of 2:1, women are more likely to be exposed to thyroid cancer than men. Diagnosing thyroid nodules is dependent on radiologists, and there may be some manual errors. To avoid these errors, computer aided diagnosis has been introduced. In the proposed work, deep learning methods were employed in the thyroid nodule classification. For classifying thyroid nodules, ultrasound image diagnosis of the nodules is chosen, as it is the most preferred and easiest method, and it also deserves various advantages over the other diagnosing methods. The aim of this work was to use deep convolutional neural networks (DenseNet) and residual networks (ResNet), as pre-trained models. These pre-trained models were obtained from keras applications, and transfer learning approach was employed in classifying thyroid nodules wherein the dataset was levied from the public access database, the Thyroid Digital Image Database. Performance metrics like accuracy, precision, sensitivity, and F1 score were captured for both ResNet and DenseNet models. A comparison between various architectures of ResNet and DenseNet was also presented. The study found that the model DenseNet201 had an accuracy of 88.89% which performed better than all other models that were trained on thyroid ultrasound images.