Abstract: As geoscience enters the era of big data, machine learning has become an emerging tool that can discover and describe complex structures and patterns of data, and is rapidly applied in the field of solid Earth geoscience. As an important subfield of machine learning, deep learning gradually learns massive amounts of data by constructing multi-level hidden layers, which can improve classification or prediction performance. However, most of machine learning models require massive amounts of data as support, which limits their widespread applications in the field of solid Earth geosciences. Transfer learning is a type of machine learning methods in the absence of adequate data, which aims to improve the performance of new tasks by using pre-trained knowledge of similar tasks in advance. By using the knowledge learned from the source domain and transferring it to the target domain, it can to some extent overcome insufficient data availability. This paper provides a brief overview of the basic concepts and categories of transfer learning, discusses the challenges faced by existing transfer learning approaches applied to geoscience by analyzing the typical cases of transfer learning in solid Earth geosciences. At present, deep transfer learning approaches have initially shown great potential in automatic identification and classification of rocks and minerals, identification of geochemical anomalies, etc. With the advantage of improving model generalization performance and avoiding overfitting, deep transfer learning approaches have broad application prospects in the field of solid Earth geosciences.