Cervical cancer remains one of the most prevalent cancers among women worldwide, with a high mortality rate among those affected. Early detection and timely treatment are crucial for prevention and reducing these mortality rates. Recent advances in deep learning, particularly transfer learning, have been employed to enhance the accuracy of cervical cancer cell classification. In this study, we present a new cervical cell dataset comprising approximately 15,000 images from Hospital A in Vietnam. We investigate various transfer learning strategies, including feature extraction and fine-tuning from scratch, for multi-class cervical cell classification. Extensive experiments were conducted using a range of image preprocessing techniques and deep learning models, including VGG19, InceptionV3, ResNet101, DenseNet201, ViT, and FasterViT. Results showed that the fine-tuning from scratch method outperforms the feature extraction method in terms of evaluation metrics.