In this article, an adaptive neural network is proposed for the tracking control problem of unknown nonlinear interconnected systems with inaccessible states and sensor delays based on dynamic surface strategy. The system has unknown nonlinearities and immeasurable states. Thus, a neural network state observer based on delayed outputs of subsystems is applied. The main difficulty in obtaining local observers’ gains is that undelayed outputs are not available. As a result, by utilizing proper Lyapunov–Krasovskii functionals in dynamic surface design procedures, the gains of local observers are given in terms of linear matrix inequalities. Then, appropriate changes in coordinates are defined using delayed outputs, observed states, and filtered virtual controls for the purpose of designing dynamic surface controllers. Subsequently, proper Lyapunov–Krasovskii functionals are introduced to deal with sensor delays and obtain control laws and stability criteria. Furthermore, the proposed decentralized control scheme can suitably conquer the decentralized tracking problem of unknown large-scale systems with sensor delays and guarantee that all the signals in the closed-loop interconnected systems be uniformly ultimately bounded. Finally, to show the effectiveness and efficiency of the proposed approach, the theoretic achievements are employed to design a controller for a double-inverted pendulum system and a cascade chemical reactor system.