Echocardiography is essential for the diagnosis and treatment of cardiovascular diseases, especially in the diagnosis of congenital heart disease. However, the interpretation of echocardiography requires the accumulation of abundant professional experience for a cardiologist, which may lead to missed diagnoses and misdiagnoses due to differences between operators. The development of artificial intelligence and its subset, deep learning, have altered this potential crisis in recent years with their rapid, accurate, and consistent nature. While the application of deep learning in echocardiography is still in its infancy, recent studies demonstrate that deep learning models can quickly obtain information by extracting samples from large databases. Moreover, growing evidence suggests that deep learning can be used for standard section recognition in echocardiography and auxiliary diagnosis of heart disease. In this review, we begin by outlining the principles of deep learning. Then, we investigate the current application of deep learning in echocardiography, underscoring its significance. Furthermore, we discuss its limitations and finally highlight future development prospects.