Abstract: With the rapid development of modern hydrological monitoring technology, massive hydrological data are obtained, which brings opportunities and challenges for deep learning in hydrology and water resources.This paper summarizes the research progress of deep learning in hydrology and water resources from three aspects: hydrological simulation, water resources management and water environment evaluation.The advantages of deep learning methods and their application difficulties are: deep learning method does not need to construct a physical model and can automatically recognize data characteristics, and has significant advantages in the problems without a clear physical mechanism; however, it faces defects such as lacking model training data, subjectivity in super-parameter determination, insufficient interpretability, inconsistency with physics laws and lacking generalization ability.At last, we propose prospects of deep learning in hydrology and water resources filed: combining deep learning and hydro-physical mechanism model to integrate classical hydrological laws and doing application researches such as transfer learning, reinforcement learning and adversarial learning to better utilize the deep learning method.