The need for Sentiment classification is a critical step in determining how people feel about a product, service, or issue. To solve the sentiment categorization problem, many natural language processing models have been developed. The majority of them, however, have concentrated on binary sentiment categorization. In this study, we tackle the fine-grained sentiment categorization task using BERT, a powerful deep learning model. Experiments show that without complicated architecture, our model outperforms other popular models in this job. In the process, we also demonstrate the utility of transfer learning in natural language processing.