Oracle Bone Inscriptions (OBI) are the earliest Chinese characters found so far, they are the treasures of Chinese traditional culture and have high research value. However, the machine recognition of OBI is a difficult problem. To address this challenge, this paper presents a new method of OBI recognition based on transfer learning and improved ResNet50 network. Firstly, the deep convolutional neural network structure of ResNet50 is used to obtain rich feature representations by learning on large general dataset, and then it is transferred to the recognition of OBI, so as to effectively extract the key features of OBI. On this basis, the parameters of ResNet50 model are fine-tuned on a small data set of a specific OBI topology to improve the recognition rate of OBI. The experiment utilized rubbed images of the original OBI as the dataset. The findings demonstrated that the proposed method achieved superior recognition accuracy, reaching 95.641% compared to alternative methods.