In this paper, a neural modified extended state observer (ESO) (NMESO) based on the compound neural orthogonal network (CONN) is proposed to improve the online estimation of the system uncertainties and the exogenous disturbances in the uncertain system. In the developed scheme, CONN is designed as the main estimator to identify the total disturbances and facilitate the estimation of the conventional ESO. The convergence of NMESO is proved in time domain and the estimation errors are shown to be bounded. Due to the orthogonality of hidden nodes and the iterative updating mechanism of CONN, the estimation accuracy, rapidity, and real-time performance of NMESO have been greatly enhanced. Furthermore, the cooperative mechanism of the CONN estimator and the conventional ESO in the developed NMESO is also studied. Finally, the effectiveness of the proposed method and the results of analysis are verified through comparative simulation experimentation. This paper depicts a promising prospect of the NMESO in application of practical engineering.