Brain source localization technology aims to identify the source of neural activity in the brain through the EEG and MEG signals on the scalp surface, which is the basis of studying the neural activity, cognitive process, and pathological function of the cerebral cortex. Its millisecond temporal resolution can effectively make up for the shortcomings of fMRI in low temporal resolution. Brain source localization contains two processes, forward problem, and inverse problem. The forward problem is to simulate the electric potential of the head surface generated by the neural source of brain activity, which is calculated by the volume conduction model, and the model is mainly built by the boundary element method, finite element method, and finite difference method. The inverse problem aims to reconstruct the distribution of current sources in the brain. The main solutions include the distributed source model and the equivalent current dipole model. But the solution to the inverse problem is not unique, and the regularization method is the classical means to resolve it, including the minimum L1 norm and the minimum L2 norm methods. Nonlinear optimization, beamforming, the Bayes approach, deep learning, and other technologies have been created in recent years to increase the accuracy of the brain source localization technique. However, due to the ill pose of the inverse problem and the errors caused by different recording methods, the number of electrodes, and head model construction in practice, the accuracy evaluation is still challenging in brain source localization, which greatly limits the practical application of brain source localization methods in neuroscience and psychology research, clinical diagnosis, and treatment. In this work, the existing brain source localization methods and analysis of the accuracy evaluation methods of brain source localization technology and its practical application in basic research and clinical diagnosis and treatment are introduced. Specifically, different recording methods, the number and density of electrodes, and the head volume conduction model all have a certain influence on the source positioning accuracy. In practice, because different inverse problem algorithms produce different source location results, this study summarizes the evaluation methods based on spatial resolution, point diffusion, and crosstalk function on the degree of source overlap among different brain source localization methods and the influence of other sources on target sources. In addition, the application of brain source localization technology in time-frequency analysis and connectivity analysis is introduced, which can help researchers better understand the connections and functions of various brain regions in cognitive activities. Currently, brain source localization technology has been used clinically in epilepsy, attention deficit, hyperactivity disorder, and other brain abnormalities or diseases. The main progresses of brain source localization technology about the abovementioned five aspects which include the process of brain source localization, the method of inverse solution, influencing factors of positioning accuracy, accuracy evaluation method, and the research and clinical application are reviewed. Furthermore, some scientific problems concerning accuracy evaluation are discussed in this paper. We hope to provide certain references and help with the development and application of brain source localization.